agentleFS
Seth090502Seth090502/osanwe-public/.agents/skills/brief/SKILL.md

brief

Skill0 starsChanged 53 days agoMITView on GitHub

Institutional-grade daily morning briefing -- today's market setup, premarket catalysts, regime. Use when starting the trading day, before the open, ahead of FOMC/CPI/earnings, on a weekend, or when the prior briefing aged past 24h. PDB BLUF/Counter/Alternative, thesis status board, net-liquidity + insider overlays, evidence-grade confidence caps, Brier-scored meta.json sidecar in Calendar/decisions/briefings/. Not a lookup in an old briefing, not a topic explainer.

  • Reads credentials

81 other projects use a version of this.

Adopted by

Most recently taken up by

Used with

Projects with it also use capture, strategy and 8 more.

Other files people use alongside this one.

Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
  • toneron2Skill

    capture

    Capture thoughts to inbox with automatic classification

    Also in 6 projects that use this

  • EpistatesSkill

    strategy

    Plan weekly or monthly X content strategy. Use when the user wants to plan their content calendar, set growth goals, build a content pipeline, or think strategically about their X presence.

    Also in 4 projects that use this

  • aleksander-dytkoSkill

    decision

    Make a product decision with full context from the vault and available sources

    Also in 3 projects that use this

  • howellsSkill

    reconcile

    Audit a project against current plugin conventions and offer to update structure.

    Also in 3 projects that use this

  • HermeticOrmusSkill

    close

    End-of-session ritual. Captures session state, updates auto-memory, writes a session summary, then exits the session cleanly.

    Also in 3 projects that use this

  • codihausSkill

    debrief

    Analyze customer requirements and deliver market research, tiered proposals, and recommendations

    Also in 3 projects that use this

The file

Show its rules and text

We split this skill into the rules it gives an agent

31 rules. 0 are common, 11 are distinctive, and 20 are very specific to this project.

  • 0

    Common

    Most of the 74 edited versions we read keep these.

  • 11

    Distinctive

    This author's own choices. Worth a look.

  • 20

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Symmetric back-linking: every wikilink in briefing related: (3-7 typical) gets reciprocal [[briefing-<date>[-HHMM]]] on the target file's …

    line 72
  2. Very specific:

    Same-day collision: if briefing-<today>.md exists AND no --replace, default to briefing-<today>-<HHMM>.md (HHMM from clock).

    line 75
  3. Very specific:

    What / So What / Now What: every portfolio mover answers all three.

    line 84
  4. Distinctive:

    Temporal anchoring: every data point has its date.

    line 89
  5. Distinctive:

    Basis-integrity caveat: never report P&L for a position whose cost basis is not positively established.

    line 93
  6. Distinctive:

    Corporate actions: verify dated split/dividend adjustments and consistent share/price basis before reporting returns or dollar impact.

    line 95
  7. Very specific:

    Quiet-day suppression: if all 7 quiet conditions from ref-monitoring-rules.md sec 2 are met, BLUF redirects to vault priorities.

    line 96
  8. Distinctive:

    If any source is empty, fall back to a generic subject for that slot (never emit an empty query).

    line 110
  9. Very specific:

    If no approved hits are available: emit Phase 0 -- no vault context retrieved with the cause and continue.

    line 133
  10. Very specific:

    U1 residency gate (orchestration economics, AGENTS.md): Run python tools/run-share.py --current-context FIRST.

    line 146
  11. Very specific:

    If --preview: stop after Phase N (composition complete in memory).

    line 219
  12. Very specific:

    If --confirm: block for the owner yes before Phase C.

    line 221
  13. Very specific:

    Theses: all 5 Atlas/concepts/investing/theses/thesis-*.md essays.

    line 234
  14. Very specific:

    Use the Agent tool with subagent_type price-fetcher.

    line 260
  15. Distinctive:

    If schema drift (any field missing): HALT.

    line 290
  16. Very specific:

    If divergent: emit Warning Problems entry with specifics.

    line 373
  17. Distinctive:

    HEALTHY: no invalidation signals; thesis confirmed by recent data

    line 390
  18. Very specific:

    Alert emission (T4 push channel one; 2026-07-04): after trigger evaluation, compute alerts_fired = newly-observed fires ONLY (condition …

    line 403
  19. Very specific:

    When the price-fetcher subagent returns top-level extended_hours_movers[] non-empty, render a dedicated table BEFORE the regular Portfolio …

    line 438
  20. Distinctive:

    Now What: NO ACTION (AH transient) / MONITOR pre-open / REVIEW THESIS at open.

    line 446
  21. Very specific:

    Quiet-day suppression interaction: if any extended_hours_movers[] entry has abs(magnitude_pct) >= 5.0, suppress the regular quiet-day BLUF …

    line 448
  22. Very specific:

    Pre-Output explicit-absence: if market_session != "regular" for ALL held positions AND extended_hours_movers[] is empty, render: …

    line 450
  23. Very specific:

    Now What: NO ACTION / MONITOR <trigger> / ADD @$X / TRIM @$X / REVIEW THESIS.

    line 459
  24. Distinctive:

    A: all theses HEALTHY, no concentration issues

    line 471
  25. Distinctive:

    If no material catalyst within 7d: skip section entirely.

    line 528
  26. Distinctive:

    Every data point has explicit date.

    line 584
  27. Very specific:

    If frontmatter confidence: exceeds applicable cap: HALT.

    line 609
  28. Very specific:

    Use the Agent tool with subagent_type vault-classifier-sweep.

    line 729
  29. Very specific:

    All four non-briefing updates MUST succeed.

    line 735
  30. Distinctive:

    After composing all target file modifications IN MEMORY but BEFORE atomic write:

    line 885
  31. Very specific:

    Write each composed file to a tmp dir under wiki/research/test-tmp/.precheck/brief-<slug>/

    line 886

“Common” is measured: kept by at least 6 in 10 of the 74 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Financial evidence contract: docs/financial-analysis-contract.md governs source quality, temporal/basis semantics and missing-data handling across harnesses. For material numeric batches, run python tools/fis/evidence.py <envelope.json> before output, or disclose manual review and any unverified semantics. Plugin interoperability: .agents/skills/finance-data/SKILL.md owns typed evidence intake and portable Finance/Data handoffs. This skill retains its financial procedure and gates. Probe callable tools at run time; installed plugin metadata alone does not establish a usable connector or source schema.

/brief -- institutional-grade morning briefing with PDB discipline (v2 final)

Take today's market state, produce an institutional-grade morning briefing that lands in the vault atomically: compose a canonical briefing file at Calendar/decisions/briefings/briefing-<date>[-HHMM].md with PDB-style BLUF + Counter + Alternative Analysis + day-regime + macro-regime + cross-asset-coherence + thesis status board + evidence-graded portfolio impact + Warning Problems + Intelligence Gaps + FOLLOWUPS:skills coordination + machine-readable meta.json sidecar; update daily note Market Pulse summary + linkback; bump hot.md last_briefing; append sessions-log entry; symmetric back-link to materially-discussed entities; commit atomically.

When to use

Daily trading-day morning use to set the day's decision frame. Weekend variant produces "Week in Review" framing with expanded calendar + scenario bar. Fed Week, earnings week, and crisis regimes auto-expand relevant sections.

Not for

  • Portfolio refresh / live-price snapshot (use /networth)
  • Thesis stress-test without a specific event (use /challenge thesis-<slug>)
  • Investment decision on a single ticker (use /invest <ticker> then /decide if action proposed)
  • Ad-hoc price check (direct script call: python tools/fetch-prices.py --equities <list>)
  • Session retrospective (use /retro)

Mode routing (Pattern 6 deterministic; invocation modes)

  • --executive / --format executive -> produce the EXECUTIVE brief format (ref-brief-executive.md): 5-10 importance-ranked narrative paragraphs with per-category portfolio effect. Replaces v2 body sections 4-19; frontmatter, BLUF block, FOLLOWUPS, and meta sidecar unchanged (sidecar gains "format": "executive"). Default remains v2 tables format.
SyntaxBehavior
/briefFull 16-phase A-P briefing; atomic commit across 5 targets + N entity back-links
/brief --previewPhases A-N in memory; render briefing + meta + audit to stdout; SKIP O-P; F11 never set
/brief --confirmBlock at Phase B for yes; block before each of the 6 Phase P writes
/brief --replaceSame-date collision overwrites briefing-<date>.md without -HHMM suffix; explicit user intent (sidecar overwrites too)
/brief --quickCompressed: Markets + Signal Dashboard + Portfolio Movers + FOLLOWUPS only; skip G/I/K/L and most of M; Phase H runs H-lite (machine-trigger evaluator + alert emission only, no prose board -- the T4 scheduled-lane alerting spine); <=200 words body; sidecar still written
/brief --refresh <path>Additive merge into existing briefing: mutate Alerts + FOLLOWUPS only; body outside byte-exact (sha256-verified); updated: bumps; append to existing meta.json web_searches + followup_skills arrays; no peripheral writes
/brief --no-peripheralPhase O briefing + sidecar only; SKIP Phase P entirely (composition test)

Execution Rules

  • Tier-A/Tier-B split (DW port, brief-vnext 2026-07-04): Phase A.7 detects topology. TOPOLOGY=dw is Tier-A and requires the harness to expose a Workflow tool -- it routes Phases C.2/D/E.0 data acquisition through the brief-research workflow (.claude/workflows/brief-research.js) per Phase A.8. TOPOLOGY=sequential is the Tier-B spine and the PERMANENT UNIVERSAL FALLBACK: every harness without a Workflow tool -- and every harness without an Agent tool at all -- executes the phases sequentially exactly as written, at full quality and with no loss of contract. No merging either way. Use WebSearch only when explicitly authorized per phase (target: 0-3 web searches per run TOTAL, distributed -- not expanded -- across DW workers; script provides everything else).
  • Harness-conditional dispatch (binds EVERY dispatching phase: A.8 wave routing, D.0 price-fetcher, J.0 institutional-positioning-scout, the A.8 lite data-integrity skeptic, P.0 vault-classifier-sweep): dispatch the named workflow/subagent WHEN the harness exposes the corresponding Workflow/Agent tool; when it does NOT, execute that phase's documented inline fallback VERBATIM as the NORMAL path -- reported topology: sequential (no Workflow tool) / mode: inline (no Agent tool), a harness capability gap, NEVER a DEVIATION. The WORK is mandatory either way; only the mechanism varies. Skipping a dispatch where the tool DOES exist remains a DEVIATION. No phase conditions its behavior on model identity.
  • F11 Phase C discipline: set .claude/state/auto-commit-disabled at Phase C, BEFORE any Edit/Write. Phase B state-transition print runs BEFORE F11 so user can abort pre-writes. Clear flag only after Phase P post-commit verification passes.
  • Path-guards mechanical. NEVER write to: private/, .raw/, _quarantine/, finance/, credentials/, .git/, .claude/hooks/, .claude/state/ (except the F11 flag + append-only alerts-delivered.log per Phase H alert emission). NEVER read .raw/, private/, finance/, credentials/, .env*, auth.json, or *.local.md into context. Account facts require current-session broker read evidence. Phase 15 portfolio refresh from /brief v1 is RETIRED -- portfolio mutations belong to /networth scope.
  • Tag vocabulary canonical: emit only topic/market-brief on briefing frontmatter (no ticker/company/thesis tags unless briefing is itself thesis-centric, rare/explicit). HALT on namespace drift.
  • Binary decisions default: thesis HEALTHY/WATCH/STRESSED/INVALIDATED; coherence coherent-on/coherent-off/divergent; followup skill EMIT/SKIP. NEUTRAL only when signal genuinely absent both directions.
  • Confidence vs Conviction are DISTINCT fields. Confidence (epistemic, 0-100%) capped by evidence-grade mix. Conviction (low/medium/high actionable) reported separately.
  • Body-preservation sha256 on every UPDATE in Phase P (daily note, hot.md, entity notes). Pre/post sha256 with strictly-additive-outside-insertion-sites assertion.
  • Symmetric back-linking: every wikilink in briefing related: (3-7 typical) gets reciprocal [[briefing-<date>[-HHMM]]] on the target file's Recent section.
  • Mid-batch failure: immediate HALT (Phase P F.halt), F11 stays on, no partial commit, structured report.
  • Commit discipline: ASCII-only title + body, Co-Authored-By SUPPRESSED (F17 verify), F14 narrow staging (explicit pathspecs only).
  • Same-day collision: if briefing-<today>.md exists AND no --replace, default to briefing-<today>-<HHMM>.md (HHMM from clock). If same-HHMM file also exists, HALT (sub-minute re-run is ambiguous intent).
  • Subagent dispatch is MANDATORY when specified by phase AND the harness exposes an Agent tool (Phase D.0 dispatches price-fetcher for live quotes; Phase J.0 conditionally dispatches institutional-positioning-scout per held position with fresh institutional signal; Phase P.0 dispatches vault-classifier-sweep pre-commit gate). Pre-emptive skip based on judgment ("script will work fine direct", "no fresh institutional signal worth checking", "audit not necessary today") is FORBIDDEN. The subagents decide applicability via their own N/A return contracts (e.g., institutional-positioning-scout returns N/A on crypto tickers gracefully). Legitimate fallback fires ONLY on (a) contract violation -- subagent return missing required fields, OR (b) actual dispatch failure -- timeout, rate limit, tool-denial, hard subagent crash. Pre-emptive skip surfaces in Phase P audit as DEVIATION (not "fallback"). Quality preserved by additive design: existing inline tools/fetch-prices.py invocation + manual portfolio impact + direct tools/vault-audit.py invocation remain intact as legitimate-failure fallbacks. --quick mode SHOULD skip Phase J.0 + P.0 dispatches (sub-30s target precludes the 5-15s dispatch overhead per call); that's a documented skip, not deviation. A harness with NO Agent tool runs those same inline paths (tools/fetch-prices.py, manual portfolio impact, direct tools/vault-audit.py) as its NORMAL path, reported mode: inline (no Agent tool) -- a capability gap, not a DEVIATION.

Quality Rules

  • BLUF first: opening sentence is a judgment with numbers + dollar impact, not a summary. Bad: "Markets are mixed". Good: "<ticker> -4.2% on China export rumors, $<amount> portfolio impact across <N> combined shares."
  • Counter line (Tenth Man, PDB ICD 203): one sentence arguing the opposite case with cited evidence. Not canned bear-vs-bull -- specific evidence-grounded dissent.
  • Alternative Analysis (PDB ICD 203): a third-order scenario distinct from BLUF and Counter. One line. The non-obvious third path that becomes obvious in hindsight.
  • Priced-In line: state market expectations + what would surprise. If nothing material today, skip the line entirely (do not pad). Each priced-in call goes into frontmatter priced_in_calls: as structured list for tomorrow's Brier scoring.
  • What / So What / Now What: every portfolio mover answers all three. Raw data without interpretation is noise. Now What is one of: NO ACTION / MONITOR <trigger> / ADD @$X / TRIM @$X / REVIEW THESIS. ADD/TRIM gets copy-paste-ready order block.
  • Dollar impacts EXACT from shares_held * (price - prev_close). Cross-account holdings combined (a name held in more than one account is summed once, never double-counted). Beta-adjusted impact for thesis-level events. Never estimate, never "about".
  • Regime classification drives expansion: Crisis/Risk-off expand portfolio + geo + alerts. Risk-on/Rotation contract alerts; expand calendar + deployment. Fed Week expand calendar + scenario bar.
  • Calibrated % confidence on forward-looking: percentages, not HIGH/MEDIUM/LOW. Capped by Pre-Output gate evidence-grade rule.
  • Inline evidence grading on material claims: [Grade A | source | date]. A = primary (SEC, IR, FRED). B = Tier 1 wire (Reuters, CNBC, AP). C = aggregator/analytical. D = sentiment. F = unverifiable. STALE (>7d) auto-downgrades one letter.
  • Temporal anchoring: every data point has its date. No "recently / today / currently" without timestamp.
  • Intelligence Gaps vs Warning Problems: Gaps = "I don't know X (impact, resolve)". Warnings = "I know X and it matters but is below trigger threshold (base rate, escalation trigger)". Distinct sections; do not conflate.
  • 3-minute rule: briefing body <=650 words. >650 = padding defect. Quick mode <=200 words.
  • CFA 3-stage geo filter: event detection -> transmission mapping -> SUPPLY (include) or SENTIMENT (skip unless FLASH). Sentiment-only events excluded without mention.
  • Basis-integrity caveat: never report P&L for a position whose cost basis is not positively established. Without positively established basis, omit P&L. Report supported current-value changes separately from realized/unrealized P&L and disclose any transfer or corporate-action caveat.
  • Cross-account holdings combined: derive overlap from current-session broker read evidence; reconcile every covered account before claiming combined exposure.
  • Corporate actions: verify dated split/dividend adjustments and consistent share/price basis before reporting returns or dollar impact; do not reuse a historical adjustment assumption.
  • Quiet-day suppression: if all 7 quiet conditions from ref-monitoring-rules.md sec 2 are met, BLUF redirects to vault priorities; FOLLOWUPS empty-case renders (no followup skills recommended).
  • ASCII-only NEW content (Pattern 22): em-dash -> --; curly quotes -> "/'; arrows -> -> / <-; <= / >=; ellipsis -> ...; euro -> EUR; NBSP -> space; middle dot -> -. Byte-scan; HALT on any byte > 127.

Phase 0: Vault context retrieval (NEW; Phase 3.6d -- runs BEFORE Phase A; read-only; runs in --preview; SKIP on --quick)

Read-only retrieval uses explicitly approved non-sensitive sources. A semantic adapter must enforce its corpus allowlist and protected-path exclusions BEFORE result text enters context; post-retrieval filtering is insufficient. Scope Phase 0 to public-research/method sources, never account histories, snapshots, dailies, logs or raw broker output. Named narrative reads in later phases retain their own scope.

0.0 -- Mode gate: if invocation includes --quick, SKIP Phase 0 (emit Phase 0 -- skipped (--quick fast path)) and proceed to Phase A. All other modes (normal, --preview, --confirm, --replace, --refresh, --no-peripheral) RUN Phase 0.

0.1 -- Derive 13 query subjects across 4 groups from current vault state:

  • PORTFOLIO (up to 4): use current broker-read positions to choose the largest known holdings for retrieval queries. If unavailable, use named thesis entities for context and mark account relevance UNVERIFIED; never consult private/ or *.local.md.
  • THESIS (5): from Atlas/concepts/investing/theses/thesis-*.md (5 essays: theme-alpha, theme-epsilon, theme-gamma, theme-beta, theme-delta). Query = "<thesis-name> thesis".
  • CATALYST (2): from Atlas/sources/investing/ref-market-calendar.md next-7-day events (FOMC / CPI / earnings / etc.). Query = "upcoming <event> catalyst impact".
  • CROSS-CUTTING (2): from the 2 most-recent wiki/research/challenges/*.md filenames. Query = "<challenge-topic> findings". If any source is empty, fall back to a generic subject for that slot (never emit an empty query).

0.2 -- Use the verified adapter on an explicit approved source set for the four groups. If no enforced safe adapter is available, use rg -n --fixed-strings -- <subject> <approved-file> ... on named entity/public-research/method files after checking their scope. Exclude protected paths, .env*, auth.json, *.local.md and personal financial snapshots before content is read. No broad wiki/, Calendar/ or wiki/investing/ query; do not assume a filter parameter enforces this boundary on an unverified external index.

0.3 -- Dedup results by path:line; tag each approved hit with portfolio, thesis, catalyst or cross-cutting. Retain semantic scores only if supplied by the adapter; lexical matches have no invented score. Store method and scope in VAULT_CONTEXT and disclose the lexical fallback when used.

0.4 -- EMIT a visible summary (MANDATORY), GROUPED by derivation source:

Phase 0 -- derived-subject vault context (N distinct paths total)
PORTFOLIO    (P hits): [score] path:line -- ~60 chars   (up to 4)
THESIS       (T hits): ...
CATALYST     (C hits): ...
CROSS-CUTTING (X hits): ...

If no approved hits are available: emit Phase 0 -- no vault context retrieved with the cause and continue. NEVER HALT on Phase 0 alone.

0.5 -- Downstream consumers (phases A-P UNCHANGED; consult VAULT_CONTEXT by group):

  • Phase D/E (markets + portfolio movers): portfolio-group context informs each mover's "So What".
  • Phase F/G (thesis status board): thesis-group context seeds HEALTHY/WATCH/STRESSED/INVALIDATED priors.
  • Phase H/I (catalysts + calendar): catalyst-group context anchors the priced-in calls.
  • Phase K/L (FOLLOWUPS:skills + Counter/AA): approved cross-cutting context surfaces cross-skill connections.

Cite any used chunk inline as (vault: path:line) -- plain text, NO wikilink syntax.

Phase A: Pre-flight (NO F11 yet)

U1 residency gate (orchestration economics, AGENTS.md). Run python tools/run-share.py --current-context FIRST. Exit 1 means the session is carrying more resident context than this run should re-pay on every turn: report the number, recommend /clear + re-invoke in a fresh session, and proceed only if the operator says to. Cost basis: cache_read is ~82% of frontier cost and equals the sum over turns of the resident prefix, so a heavy run started at 500K residency costs multiples of the same run started fresh. This is a RECOMMENDATION, not a HALT -- never block work the operator asked for; make the price visible before it is paid.

A.1 -- Parse args: --preview, --confirm, --replace, --quick, --refresh <path>, --no-peripheral. Validate combinations (e.g., --refresh requires path; --replace and --quick are exclusive of --refresh).

A.2 -- Resolve today's ISO date from system clock (date +%Y-%m-%d); never from prior session memory or context.

A.3 -- F11 collision check: if .claude/state/auto-commit-disabled exists, HALT: "F11 already set by another skill. Either another skill is running concurrently or a prior crash left an orphaned flag. Manually rm .claude/state/auto-commit-disabled to clear, then re-invoke."

A.4 -- Broker coverage: discover available read tools and obtain current-session account/asset evidence. Record read time and covered scope; unavailable coverage sets BROKERAGE_WARN = "UNVERIFIED: <missing scope>" for the header and Intelligence Gaps. Public-market briefing work continues, but affected dollar impact, P&L and whole-book claims remain unavailable; no saved-file fallback.

A.5 -- Same-day collision detection: if Calendar/decisions/briefings/briefing-<today>.md exists AND no --replace, default output to briefing-<today>-<HHMM>.md (HHMM from clock). If same-HHMM file also exists, HALT (sub-minute re-run ambiguous; ask the owner).

A.6 -- Validate tools/fetch-prices.py exists; HALT if missing (script is load-bearing for Phase D).

A.7 -- Topology detection (mirror of /invest A.7): TOPOLOGY=dw iff the Workflow tool is present in the session tool surface; else TOPOLOGY=sequential. Ambiguity -> sequential. capability-detect.sh is ADVISORY only -- the tool-surface check is authoritative. ALSO record AGENT_TOOL = present|absent (does this harness expose an Agent/subagent tool? dw implies present): AGENT_TOOL=absent routes every dispatching phase (D.0, J.0, the A.8 lite skeptic, P.0) to its documented inline path as the NORMAL path, reported mode: inline (no Agent tool). Record ORCHESTRATOR_MODEL (passive logging only -- NO behavior branches on model identity, per Topology & Data Doctrine). --quick FORCES sequential (sub-30s target precludes workflow spin-up; documented skip, not deviation). Frontmatter + sidecar gain topology: + orchestrator_model: (passive).

A.8 -- Wave routing:

  • TOPOLOGY=sequential (Tier-B; every harness with no Workflow tool): Phases C.2/D/E.0 run verbatim as written below -- zero behavioral difference; the PERMANENT UNIVERSAL FALLBACK, reported topology: sequential (no Workflow tool) and never a DEVIATION. With AGENT_TOOL=present the D.0 / J.0 / P.0 Agent dispatches still fire inside this spine; with AGENT_TOOL=absent each falls through to its documented inline path (mode: inline (no Agent tool)).
  • TOPOLOGY=dw (Tier-A; ONLY where the harness exposes a Workflow tool): invoke the brief-research workflow (.claude/workflows/brief-research.js; Workflow tool, {scriptPath}) with args {equities, crypto, held_tickers, thesis_slugs, run_date, token_budget}. ONE wave, concurrency <=6, of independent READ-ONLY acquirers: (1) price-fetcher agentType (existing contract incl. extended-hours fields; replaces the Phase D.0 Agent-tool dispatch), (2) FRED macro worker (Phase E.0 series set incl. net-liquidity rows; availability-guarded N/A return; also carries the Phase J.6 openinsider pair), (3) entity+challenge recency worker (git log + targeted Reads; returns Phase C.2 recency summaries), (4) continuity worker (last-3 briefings parse + brier-ledger read + python tools/score-outcomes.py --dry-run --json; returns the calibration block), (5) an optional personal worker that is not included in this copy. Main loop RETAINS: current-session broker reconciliation (workers receive only the minimum permitted evidence needed for their task; no protected-file reads), regime CLASSIFICATION (tables unchanged; workers only fetch inputs), all composition phases F-N, all writes O-P. Convergence gate identical to /invest: N/A = success; missing fields -> one re-dispatch -> documented inline fallback; pre-emptive skip = DEVIATION in Phase P audit.
  • Token budget: BRIEF_DW_TOKEN_BUDGET env var, default 150K (PLACEHOLDER -- recalibrate after 3 instrumented runs, same protocol as INVEST_DW_TOKEN_BUDGET). Main loop reads env and passes args.token_budget (the workflow sandbox has no process.env).
  • Adversarial wave (lite; conditional -- NO standing Wave-3): a single data-integrity skeptic (read-only Agent dispatch, ~15-20K tokens) fires ONLY when (a) any ADD/TRIM order block is emitted, (b) any thesis shifts to STRESSED/INVALIDATED, or (c) portfolio_health is D/F. It re-derives the dollar math (shares x (price - prev_close)), checks order-block prices against quotes, and validates the shift's trigger citation. Fail-closed: refutation -> fix data or downgrade the line to MONITOR + Warning Problems entry; never override (Wave-3 rule). Expected fire rate <20% of mornings. Fires in BOTH topologies (sequential included) since it needs the composed draft; runs between Phase N gate and Phase O writes. Harness-conditional: on AGENT_TOOL=absent the main loop performs the SAME re-derivation inline -- recompute shares x (price - prev_close), re-check every order-block price against the Phase D quotes, re-verify the shift's trigger citation -- with identical fail-closed semantics, reported skeptic: inline (no Agent tool); never a DEVIATION.

Phase B: State-transition model (print only, BEFORE F11)

Emit to stdout the planned state transition. User can abort here pre-F11; zero writes will have occurred.

## /brief v2 -- planned state transition

MODE:                [normal|preview|confirm|quick|refresh|no-peripheral]
DATE:                <YYYY-MM-DD>
OUTPUT:              Calendar/decisions/briefings/briefing-<date>[-HHMM].md
COLLISION:           [none|-HHMM variant|--replace overwrite]
MARKET STATUS:       <resolving in Phase D from script last_trading_day>
BROKER COVERAGE:     <read time + covered scope|UNVERIFIED: missing scope>

READS:
  - Current-session broker read evidence (covered accounts/assets + time)
  - Atlas/sources/investing/ref-{monitoring-rules,portfolio-doctrine,geopolitical-framework,market-calendar,macro-landscape}.md
  - Atlas/concepts/investing/{macro-outlook,watchlist,investing-research-log}.md
  - Atlas/concepts/investing/theses/thesis-*.md (5 theses; for invalidation triggers)
  - wiki/hot.md (full body)
  - wiki/entities/tickers/*.md modified <7d (N files; surface entity-level changes)
  - wiki/research/challenges/*.md modified <14d (N files; surface invalidation verdicts)
  - Calendar/decisions/briefings/ last 3 (continuity audit; Brier scoring)
  - Calendar/daily/<today>.md (Phase P pre-write race handling)

WRITES (ATOMIC):
  1. Calendar/decisions/briefings/briefing-<date>[-HHMM].md           [NEW]
  2. Calendar/decisions/briefings/briefing-<date>[-HHMM]-meta.json    [NEW; sidecar]
  3. Calendar/daily/<today>.md (## Market Pulse + linkback)           [UPDATE; sha256-gated outside section]
  4. wiki/hot.md (last_briefing bump + additive pending merge)        [UPDATE; sha256-gated]
  5. Calendar/decisions/sessions-log.md (append entry)                [UPDATE; strictly-additive]
  6. wiki/entities/tickers/<T>.md for each entity in related: (3-7)   [UPDATE; sha256-gated additive]
  7. Calendar/decisions/briefings/brier-ledger.json (append calls; resolve matured) [UPDATE|NEW; machine-owned JSON]

SCRIPT: python tools/fetch-prices.py --equities <list> --crypto <list>

PROCEED? Abort now = pre-F11, zero writes.

If --preview: stop after Phase N (composition complete in memory); render briefing + meta + audit to stdout; SKIP O-P; F11 never set; exit clean.

If --confirm: block for the owner yes before Phase C; also block before each Phase P write.

Else: autonomously proceed.

Phase C: F11 set + Context Load

C.1 -- mkdir -p .claude/state && touch .claude/state/auto-commit-disabled. Single flag covers entire invocation.

C.2 -- Parallel context load (READ-only):

  • Portfolio: obtain current {ticker, shares, cost_basis, account} from available broker read tools. Reconcile account coverage and overlapping positions; unavailable basis/quantities stay UNVERIFIED. Apply corporate-action normalization only when supported by dated evidence. Never read private/ or *.local.md.
  • Constraints/caveats: current-session user instructions and dated authoritative corporate-action/basis evidence; unresolved basis means no P&L.
  • Refs: Atlas/sources/investing/ref-{macro-landscape,monitoring-rules,portfolio-doctrine,geopolitical-framework,market-calendar}.md. Macro outlook + watchlist + investing-research-log from Atlas/concepts/investing/.
  • Theses: all 5 Atlas/concepts/investing/theses/thesis-*.md essays. Extract per-thesis invalidation-triggers list. HALT Phase H if any thesis file missing (structural gap; ask the owner).
  • hot.md: full body. Extract Last Session + Pending Items + Active Context. Capture last_briefing: ISO for continuity context.
  • Entity recency: git log --since='7 days ago' --name-only -- 'wiki/entities/tickers/*.md' | sort -u. For each, Read; extract any thesis-status shifts in body + recent Financial signals entries. Surface in Phase H input.
  • Challenge recency: git log --since='14 days ago' --name-only -- 'wiki/research/challenges/*.md' | sort -u. For each, Read; extract invalidation verdict.
  • Continuity audit (brier-ledger wiring, brief-vnext 2026-07-04): Read the WHOLE machine-owned ledger Calendar/decisions/briefings/brier-ledger.json (single source of truth for scorable calls; NOT just last-3) plus the last 3 briefings (Glob Calendar/decisions/briefings/briefing-*.md sorted; take 3 most recent) for narrative context. For each open ledger call, classify CONFIRMED/SURPRISED/PENDING against today's market state. Brier score BS = mean((forecast_prob - outcome)^2): n>=5 resolved calls in rolling 30d -> full display; n>=3 -> provisional display (Brier 0.XX (provisional, n=3)); below 3 -> null. Carry forward prior_brier_score_30d (sourced from the ledger) into today's frontmatter; emit calibration footer ("Yesterday: priced-in CPI 3.0-3.3% base 55%; actual 3.1% -> Base HIT"). Ledger absent -> create it in Phase P with today's calls (first-run bootstrap; null Brier). ALSO run python tools/score-outcomes.py --dry-run --json (read-only; offline-safe; absent script -> skip silently) and surface brier_new / brier_shadow / n_scored as one footer line: Invest-side calibration: Brier(new)=0.XX vs shadow 0.XX over n=N.

C.3 -- Build internal checklist:

  • Holdings list with combined-account totals
  • Equity/ETF tickers + crypto tickers for Phase D script call
  • Per-thesis invalidation criteria (5 lists)
  • 7-day catalyst set (FOMC/CPI/earnings/crypto-reg)
  • position sensitivity matrix (positions resolved at run time)
  • hot.md pending-items count + oldest age
  • Entity recency summary; challenge recency summary
  • Prior-briefing continuity signal (Brier carry-forward + outstanding priced-in calls)

C.4 -- Today's daily note: do NOT compute daily_before_sha256 here. Recompute immediately before Phase P Update 1 to handle UserPromptSubmit hook race (hook may append to ## Log between Phase C and Phase P).

Phase D: Live Data Fetch

Phase D.0: DELEGATED dispatch to price-fetcher (PREFERRED path; Phase C wiring 2026-05-02)

MANDATORY (per Execution Rules; SKIPPED in --quick mode per documented exception): dispatch first, no pre-emptive skip outside --quick. Subagent wraps tools/fetch-prices.py and falls back to per-ticker WebSearch on script failure internally.

Harness-conditional: dispatch price-fetcher when the harness exposes an Agent tool; when it does NOT, the D.1/D.2 inline python tools/fetch-prices.py invocation IS the normal path (reported mode: inline (no Agent tool), not a DEVIATION). On TOPOLOGY=dw the price-fetcher leg inside the brief-research workflow satisfies this dispatch.

Use the Agent tool with subagent_type price-fetcher. Pass input: {equities: [<EQUITIES from D.1>], crypto: [<CRYPTO from D.1>]}. Expected return: structured JSON quotes map with timestamp, quotes (per-ticker price, currency, source, freshness, caveat?, plus extended-hours fields: extended_hours_last, extended_hours_change_pct, extended_hours_volume, extended_hours_last_timestamp, market_session, ah_source), top-level extended_hours_movers[] aggregation, failures array.

Validate return: every held ticker present in quotes OR failures; any transferred-basis entry has a caveat: "no_pnl" field; every quote has market_session field (pre-market | regular | after-hours | closed); extended_hours_movers is always a list (empty when no AH movers above AH_MOVER_THRESHOLD_PCT default 3.0%).

On contract violation OR dispatch failure: fall through to D.2 inline python tools/fetch-prices.py invocation. Surface fallback in Phase P audit report.

On dispatch success: skip D.2 inline script invocation; proceed directly to D.3 schema validation using subagent's returned JSON.

D.1: Inline data prep (always runs)

D.1 -- Build ticker lists from Phase C portfolio + watchlist:

  • EQUITIES = [<dedup equity + ETF tickers across all covered accounts + watchlist>]
  • CRYPTO = [<dedup crypto tickers without -USD suffix>]

D.2 -- Execute (Windows: python, never python3):

python tools/fetch-prices.py \
  --equities "<comma-separated EQUITIES>" \
  --crypto "<comma-separated CRYPTO>"

D.3 -- Parse JSON. Assert schema presence:

  • indices: SPX, NASDAQ, DOW, VIX (with term_structure), 10Y, DXY, GOLD, OIL
  • equities[]: price, prev_close, change_pct, rsi14, ma50, ma200, beta, short_ratio, pct_from_52wk_high, next_earnings
  • crypto[]: price, change_pct
  • signals: oversold/overbought/below_ma50/below_ma200/earnings_within_10d/high_short_ratio
  • market_status: open|closed|pre_market|post_market
  • last_trading_day: ISO date

If schema drift (any field missing): HALT. Do not infer.

D.4 -- Market-hours branching:

  • closed + weekend (Saturday/Sunday) -> header Weekend Brief -- data as of <last_trading_day> close; skip Overnight Earnings; expand Calendar + Scenario Bar; include "Week in Review" one-liner under BLUF
  • closed + weekday -> Holiday Brief -- data as of <last_trading_day> close; treat last_trading_day as frozen
  • pre_market -> note "Pre-market; prices as of last close"; ONE web search authorized for futures
  • open / post_market -> normal flow

D.5 -- Fallback: if errors array contains >50% of EQUITIES tickers, set SCRIPT_FALLBACK = true; web-search for missing prices; annotate "Script fallback -- lower precision" in Intelligence Gaps; force confidence cap to 60% in Phase N.

D.6 -- Compute script_output_hash = sha256(stdout) for meta.json.

Phase E: Regime Detection (day + macro, two-tier)

Phase E.0: FRED macro fetch (ADD-NOW 2026-06-05; authoritative regime inputs; availability-guarded)

Before classifying, fetch the regime input set from the FRED MCP per ref-macro-data-sources.md (sibling of this file): mcp__fred__fred_get_series for DGS10, T10Y3M, T10Y2Y, BAMLH0A0HYM2 (HY OAS), VIXCLS, UNRATE (latest + 2-3 prior for trend), CPILFESL (units=pc1), FEDFUNDS, WALCL + RRPONTSYD + WTREGEN (net-liquidity composite, 4wk trend), DTWEXBGS (broad dollar index; the Grade-A macro-regime dollar input -- script DXY stays in the Markets one-liner). Tag provenance FRED:<series_id> asof <obs_date> + freshness. The classification TABLES below are UNCHANGED -- FRED only replaces the prior web-search source for the credit/curve/unemployment inputs. Availability-guard: on MCP absence/error, SKIP FRED silently and fall back to the Phase D script 10Y/VIX + ONE web search for HY OAS (contract byte-stable; NEVER HALT). Additive DATA-SOURCING only -- does NOT change any regime classification rule or briefing verdict.

Day regime (tactical, 6-class). Source VIX from FRED VIXCLS + Phase D term-structure, credit from FRED BAMLH0A0HYM2 (bps); the prior web-search-for-OAS is now the FRED fallback only (VIX > 25 = Risk-off regardless of OAS):

RegimeVIXTermCreditSection action
Crisis>30BackwardationHY OAS >500bpsExpand portfolio/geo/alerts
Risk-off20-30Flat/BackwardationOAS wideningExpand portfolio movers + geo
EarningsanyanynormalExpand earnings
Fed WeekanyanysensitiveExpand calendar + scenario bar
Risk-on<16ContangotighteningContract alerts; expand deployment
Rotation16-20ContangostableNormal; note flow direction

Report classification + which markers matched (Pattern 11 auditable): VIX 24.3 [contango], OAS +12bps [widening], XLU/XLP rel-strength +0.8% -> Risk-off.

Macro regime (cycle, 4-class) from yield curve (FRED T10Y3M/T10Y2Y) + unemployment trend (FRED UNRATE latest + priors) + credit spreads (FRED BAMLH0A0HYM2) + ISM if available (ISM is NOT on FRED per cap-matrix C6 -- web only; curve+unemployment+credit carry the classification):

Phase10Y-3MUnemploymentCreditLeadership
Early-cycleSteepFallingTight/tighteningCyclicals, small-caps
Mid-cycleNormalStable lowStableBalanced
Late-cycleFlat/invertedStable lowTight but wideningDefensives, quality
ContractionRe-steepeningRisingWide/wideningBond proxies, gold

Both regimes go to frontmatter (regime: + macro_regime:). Transition flag: if macro_regime differs from prior-3-briefings mode, mark macro_regime_transition: true and surface in Warning Problems.

Phase F: Signal Dashboard + Markets one-liner

Markets one-liner (single line, all from Phase D script):

S&P $X,XXX (+/-X.X%) | Nasdaq $XX,XXX (+/-X.X%) | 10Y X.XX% [+/-Xbps] | BTC $XX,XXX (+/-X.X%) | DXY XX.X | VIX XX.X [cont|bkwd] | Oil $XX (+/-X.X%) | Gold $X,XXX
NetLiq $X.XXT [4wk +/-$XXB] | Broad$ XXX.X

(NetLiq line from FRED WALCL - RRPONTSYD - WTREGEN per ref-macro-data-sources.md; reported context only -- NOT a regime axis; a legal Warning Problems input when 4wk delta < -$150B. Skip the line silently on FRED absence.)

Signal Dashboard (compact table over held equity/ETF positions, 8-12 lines max including header):

| Ticker | Price | Chg% | AH/PM | RSI | vs 50MA | vs 52wk H | Beta | Signal |

The AH/PM column renders when ANY held position has market_session != "regular" (i.e., AH session active globally) OR when individual ticker has non-null extended_hours_change_pct. Format:

  • +X.X% AH (after-hours change_pct, signed; AH session)
  • +X.X% PM (pre-market change_pct, signed; PM session)
  • $X (untraded) (market_session in AH/PM but extended_hours_volume == 0; rare; signal = ticker has AH session but untraded that minute)
  • - (regular session OR market closed; column dropped if NO held position has AH/PM data globally)

Signal logic:

  • RSI <30 AND price > 200MA -> OVERSOLD (cite ~73% bounce base rate when regime is non-bear)
  • RSI >70 -> OVERBOUGHT
  • price < 50MA AND price < 200MA -> WEAK
  • earnings <=10 trading days -> PRE-EARN Nd
  • short_ratio >5 -> HIGH-SHORT
  • ah_movers ticker (i.e., abs(extended_hours_change_pct) >= 3%) -> AH-MOVER (combines with other signals: AH-MOVER + PRE-EARN 2d)
  • multi-signal -> combine: WEAK + PRE-EARN 7d
  • else -> NEUTRAL

Section budget: 8-12 lines. If >8 held positions, only render non-NEUTRAL rows; else render all (no hiding).

Phase G: Cross-Asset Coherence Check

Classify session as one of three:

  • coherent-risk-on: equities up, credit tight/tightening, rates up (growth bid), USD down, gold down, VIX down, BTC up
  • coherent-risk-off: equities down, credit wide/widening, rates down (flight to safety), USD up, gold up, VIX up, BTC down
  • divergent: any material dissent between signals

If divergent: emit Warning Problems entry with specifics. Example: Equity up, credit widening, gold up -- dissent between equity rally and risk signals; idiosyncratic sector move OR regime transition in progress.

Frontmatter field: cross_asset_coherence: <coherent-on|coherent-off|divergent>. Divergent sessions historically precede regime shifts; flagging is the point.

Phase H: Thesis Status Board (NEW v2; binary per-thesis)

For each of the 5 active theses (theme-alpha, theme-beta, theme-gamma, theme-delta, theme-epsilon), evaluate status against invalidation triggers from Atlas/concepts/investing/theses/thesis-<slug>.md (loaded Phase C):

ThesisStatusEvidence
theme-alphaHEALTHY / WATCH / STRESSED / INVALIDATED<specific trigger state, one line; when a capex-cut exit-ladder leg is armed, append exit-ladder: <tier> per ref-portfolio-doctrine>
theme-beta......
theme-gamma......
theme-delta......
theme-epsilon......

Binary per Pattern 7. Definitions:

  • HEALTHY: no invalidation signals; thesis confirmed by recent data
  • WATCH: signal present but below trigger threshold
  • STRESSED: trigger threshold crossed, pre-invalidation
  • INVALIDATED: thesis premise broken

Machine-trigger evaluator (thesis machine-triggers schema v1; ratified Block A 2026-07-04): when a thesis essay carries a triggers: frontmatter block, evaluate BEFORE judgment-layer prose reasoning:

  • Parse triggers: (kill/red/amber/manual rows; schema v1 per proposal-thesis-machine-triggers-2026-06-10). Evaluate every kill/red/amber whose source resolves from data already in hand (Phase D script quotes, E.0 FRED, J.6 openinsider). manual: rows are never evaluated.
  • Status FLOOR derivation (floors combine with prose judgment; judgment may worsen but never improve below the floor): any kill fired -> STRESSED minimum (INVALIDATED stays a human call via /challenge); >=1 red fired -> WATCH minimum + Warning Problems entry; >=1 amber fired -> WATCH candidate.
  • Render per-trigger state in the Status Board Evidence cell (e.g. <trigger-id>: FIRED <date>, /decide pending) + a staleness line per thesis: manual: N unevaluated (oldest review <date>).
  • Write-back of fired: is Pattern-20-gated: /brief NEVER edits Atlas. A newly-observed fire emits a FOLLOWUPS:skills line (/decide -- trigger <id> fired; ratify fired-date write-back) and the write lands in the confirmed session. Evaluation itself is read-only.
  • Back-compat: thesis essay without a triggers: block -> prose-judgment evaluation exactly as before (zero behavior change).
  • GATE-T adjudication (judgment-gates kit, 2026-07-06): manual: rows and evidence-interpretation triggers (source: manual / window: earnings_window) are never machine-evaluated. When the prose judgment layer proposes a thesis-status change (either direction) resting on such a trigger, a same-day GATE-T sheet is REQUIRED: run the GATE-T procedure per .agents/skills/gate/ref-gate-tables.md (fill evidence markers with provenance, python tools/gate-eval.py --compute -- never judge the verdict -- write the sheet to wiki/research/gates/, --check, registry row), and cite the sheet in the Evidence cell. FIRED backs the status floor (the fired: write-back stays Pattern-20-gated via /decide FOLLOWUP; the sheet is its evidence record). INSUFFICIENT-EVIDENCE -> status unchanged + evidence-collection EOD row. NOT-FIRED -> status unchanged; the sheet prevents re-litigating the same evidence daily. N/A in --quick (H-lite has no prose judgment layer). Enforced by HALT item 22a.

Alert emission (T4 push channel one; 2026-07-04): after trigger evaluation, compute alerts_fired = newly-observed fires ONLY (condition evaluates true AND no fired: date recorded in the essay frontmatter; recorded-fired rows re-render in the board but NEVER re-alert -- alarm-fatigue guard). For each entry {id: "<thesis>-<trigger-id>-<YYYY-MM-DD>", trigger, thesis, class, detail, push}:

  • Send PushNotification: one line, trigger id + thesis + observed value vs threshold (+ dollar exposure when in hand). Tool unavailable/failed -> push: "failed", continue (never HALT the briefing on the alert leg). Interactive sessions may return not-sent/terminal-active (harness dedup; the alert reaches the user in-session) -> push: "skipped-active", delivery-equivalent.
  • Append <ISO-8601> <id> push=<requested|failed|skipped-active> <detail> to .claude/state/alerts-delivered.log (the X75 delivery receipt; tools/open-loops.py matches on the id substring against this file).
  • alerts_fired lands in meta.json (schema v2 field). Empty list when no new fires: no push, log untouched.

H-lite (--quick mode): --quick runs ONLY the machine-trigger evaluator + alert emission from this phase -- no prose Thesis Status Board body section; thesis_statuses frontmatter + sidecar still emitted (floor-derived where triggers resolve, prior-briefing carry-forward elsewhere, carry-forward rows suffixed (carry)). Triggers whose source data is not in hand under the quick fast path (E.0 FRED skipped, J.6 skipped) render not-evaluated (quick) and are NEVER treated as fired. This is the scheduled pre-market lane's alerting spine (tools/run-morning-brief.cmd).

Frontmatter: thesis_statuses: {theme-alpha: HEALTHY, theme-beta: WATCH, ...}.

Action: any thesis shifting HEALTHY -> WATCH (or worse) emits /challenge thesis-<slug> to FOLLOWUPS:skills + Warning Problems entry.

Failure: all 5 thesis files missing -> HALT Phase H with explicit error; ask the owner (structural gap).

Phase I: Overnight Earnings + Geopolitical (conditional; skip when absent)

Overnight Earnings: skip section entirely if no held/watchlist position reported. For each reporter:

  • EPS actual vs consensus, surprise %
  • Revenue actual vs consensus, surprise %
  • Guidance: raised/maintained/lowered vs prior + vs consensus
  • Pre-market reaction %
  • Thesis implication: CONFIRM or CHALLENGE (NEUTRAL only when truly absent both ways)
  • PEAD direction (5-60d post-earnings drift magnitude per academic literature)
  • Peer-constellation signal (e.g., "SK Hynix beat -> positive PEAD signal for MU")

Geopolitical: CFA 3-stage filter from ref-geopolitical-framework.md:

  1. Event detection
  2. Transmission mapping (EVENT -> POLICY -> MARKET -> SECTOR -> POSITION)
  3. Classification: SUPPLY (include) or SENTIMENT (skip unless FLASH-level)

Cross-reference position sensitivity matrix (positions resolved at run time). Skip section entirely if nothing survives both filter and threshold.

Phase J: Portfolio Impact

Extended-Hours Movers (subsection; renders BEFORE regular movers when extended_hours_movers[] non-empty)

When the price-fetcher subagent returns top-level extended_hours_movers[] non-empty, render a dedicated table BEFORE the regular Portfolio Impact table:

| Ticker | Session | AH Move | $ Impact (regular close -> AH last) | So What | Now What |

  • Session: pre-market | after-hours
  • AH Move: signed magnitude_pct from quote (e.g., +5.4% AH); include last_timestamp parenthetical if recent (e.g., +5.4% AH (19:59 ET))
  • $ Impact: shares_held * (extended_hours_last - regular_close) EXACT; cross-account combined only from verified quantities; current-value change is distinct from P&L and carries basis caveats
  • So What: thesis CONFIRM/CHALLENGE/NEUTRAL using AH price as evidence; cite specific invalidation trigger if CHALLENGE
  • Now What: NO ACTION (AH transient) / MONITOR pre-open / REVIEW THESIS at open. Order blocks NOT embedded for AH movers (AH liquidity thin; defer to regular-session execution unless user explicitly requests AH limit order)

Quiet-day suppression interaction: if any extended_hours_movers[] entry has abs(magnitude_pct) >= 5.0, suppress the regular quiet-day BLUF redirect (AH-quiet != regular-quiet); render full briefing.

Pre-Output explicit-absence: if market_session != "regular" for ALL held positions AND extended_hours_movers[] is empty, render: [Extended-hours session active: no material movers above 3.0% threshold across held positions].

Regular session table (positions moving >1%):

| Ticker | What ($ impact, exact) | So What (thesis CONFIRM/CHALLENGE) | Now What |

  • Dollar impact: shares_held * (price - prev_close) EXACT. Cross-account positions combined.
  • Beta-adjusted impact for thesis-level events: position_impact * beta.
  • So What: thesis status; if CHALLENGE, cite specific invalidation trigger from ref-portfolio-doctrine.md.
  • Now What: NO ACTION / MONITOR <trigger> / ADD @$X / TRIM @$X / REVIEW THESIS. If ADD or TRIM, embed copy-paste-ready order block:
ORDER (confirm at broker):
SELL <N> shares <TICKER> @ $<price> LIMIT

Crypto movers carrying the no_pnl caveat report current value-delta only: <SYMBOL> $X.XX (+X.X%): ~$X delta on N tokens. P&L NOT REPORTED -- cost basis not established.

Any mover with a recent split: report pre-split + post-split prices; share count uses post-split.

Portfolio Health grade (A-F):

  • A: all theses HEALTHY, no concentration issues
  • B: 1 thesis WATCH OR minor concentration drift
  • C: 2+ theses WATCH OR notable concentration
  • D: any thesis STRESSED
  • F: any thesis INVALIDATED OR crisis-level drawdown

Frontmatter: portfolio_health: <A|B|C|D|F>.

Phase J.6: Insider overlay (openinsider; NEW brief-vnext 2026-07-04)

  • TOPOLOGY=dw: folded into the Wave-1 FRED/macro worker as a second tool-call pair. Sequential: two direct MCP calls.
  • mcp__openinsider__top_buys + mcp__openinsider__top_sells (window: trailing week), INTERSECT against held + watchlist tickers only. Render 0-3 lines max inside Portfolio Impact: INSIDER: <TICKER> cluster buy $X.XM by N insiders [Grade B | openinsider | date] -> <one-phrase So What>. Empty intersection -> render nothing (quiet-day discipline; no padding).
  • Availability-guard: MCP absent/error -> skip silently; note in Intelligence Gaps only if a held ticker had a known pending Form-4 storyline. --quick skips. This is the /brief-scale sibling of /invest Phase J-bis -- deliberately NOT dispatching institutional-positioning-scout daily (cost discipline; the scout remains /invest's depth tool).

Phase K: Calendar (7d) + Pre-Earnings Monitor + Scenario Bar

Calendar (7 days): from ref-market-calendar.md + script earnings_within_10d + 0-1 web search for gaps. Each event:

  • Date + name
  • Conditional framing: If hot X -> implication A; if soft Y -> implication B
  • Historical one-liner: avg move last 3 occurrences if data available

Include FOMC, CPI/PPI/jobs, held/watchlist earnings, corporate-action dates, crypto regulatory dates.

Pre-Earnings Monitor: for each held with earnings <=10 trading days (script signals.earnings_within_10d):

  • Countdown: [TICKER] earnings in X trading days (<date>)
  • Consensus: EPS $X.XX, Revenue $X.XB
  • 30-day analyst revision trend (up/down/flat)
  • Historical pattern from ref-portfolio-doctrine.md (e.g., a name's historical post-earnings pattern)
  • Implied move from options if findable
  • PEAD peer signal direction

Scenario Bar (ONLY for highest-conviction catalyst within 7d; SINGLE, not multiple):

SCENARIO BAR: CPI Mar (2026-04-15)
Bull (25%): <3.0% -> 10Y to 4.15%, <fund> +$XX
Base (55%): 3.0-3.3% -> muted, +/-$XX
Bear (20%): >3.3% -> 10Y to 4.50%, <fund> -$XX
Expected Value: +$XX

Probabilities sum to 100%. Dollar impacts from actual share counts. EV = probability-weighted sum.

Each scenario writes to frontmatter priced_in_calls: as structured list:

priced_in_calls:
  - event: "CPI Mar 2026"
    date: 2026-04-15
    bull_prob: 25
    base_prob: 55
    bear_prob: 20
    trigger_level: "3.0-3.3% YoY core"

This is the input for tomorrow's Brier scoring (Phase C continuity audit).

If no material catalyst within 7d: skip section entirely.

Phase L: Intelligence Gaps + Warning Problems

Intelligence Gaps (known unknowns; 1-3 bullets max). Format: GAP: <unknown>. Impact: <why it matters>. Resolve: <how to fix>.

Skip section entirely on quiet days (suppression triggers met). Only emit "No material intelligence gaps." when truly absent (not as filler).

Warning Problems (IC convention; NEW v2). Distinct from Gaps:

  • Gaps = "I don't know X"
  • Warnings = "I know X and it matters, but it's below the trigger"

Emit Warning Problems for:

  • Divergent cross-asset coherence (from Phase G)
  • Macro regime transition (from Phase E)
  • Thesis moving HEALTHY -> WATCH (from Phase H)
  • Concentration drift past doctrine trigger (theme-alpha >60% amber / >70% red [interim 50% phase-in]; any other thesis >40%; any single name >30% -- per ref-portfolio-doctrine.md)
  • Any Phase D script error affecting >1 held position

Format: WARNING: <signal>. Base rate: <historical context>. Trigger: <what would escalate to action>.

Phase M: Optionality Map + Today's Focus + FOLLOWUPS:skills

Optionality Map (concrete deployable optionality):

  • Available cash: $X per covered account (current-session broker read)
  • Positions at action zones: <ticker at $X vs entry zone $Y-Z from watchlist.md>
  • Pre-registered watch zones (DISPLAY ONLY, 2026-07-30): render the watch_zone: block from each name's LATEST wiki/investing/analyses/<ticker>-analysis-*.md frontmatter as one line each -- <TICKER> watch zone <low>-<high> (set <date>, expires <date>); spot <px> <inside|above|below>. No evaluator, no alert, no artifact: /brief displays the zone, it never fires on it (contract: ref-analysis-template 6.1.3).
  • Pending decisions from hot.md: count + oldest age
  • Yesterday's commitments: X/Y completed from prior daily-note ## Commitments checkbox count; list any carried forward

Today's Focus (1-3 specific actions tied to what THIS briefing surfaced). Each with time-to-decision marker: WITHIN 30 MIN OF OPEN / BEFORE FOMC 2PM / EOW. No generic advice.

FOLLOWUPS:skills block (Pattern 18 coordination; analog to /invest INGEST:claims):

<!-- FOLLOWUPS:skills -->
- /invest <ticker> -- export-control signal pressures the theme-alpha thesis (trigger: WITHIN 48H if confirmed by Commerce Dept)
- /challenge thesis-theme-alpha -- theme-alpha effective exposure <pct> approaching the interim amber (trigger: EOW)
- /networth -- portfolio value delta >5% since last snapshot; refresh recommended
- /decide -- a trim trigger fired; structured decision record if action taken
- /retro -- queue at session end if substantive trades/decisions emerge
<!-- /FOLLOWUPS:skills -->

Empty case: render (no followup skills recommended) between tags. Do NOT emit a skill without a concrete one-line rationale tied to briefing body. Downstream /retro must be able to parse this block.

Phase N: Pre-Output HALT Gate (22-item)

Assert each item below before any Write. Any failure -> HALT, report which item failed, no writes, no commit, F11 stays on. Gate runs in memory; Phase O does the writing.

  1. BLUF is a judgment with numbers + dollar impact, not a summary
  2. Counter line present (Tenth Man; specific evidence cited, not canned opposite)
  3. Priced-In line states expectations + what would surprise, OR explicitly skipped (nothing material)
  4. Alternative Analysis present -- a third-order scenario with evidence (not just bear vs bull)
  5. Dollar impacts computed from shares_held * (price - prev_close) exact; no estimates
  6. Signal Dashboard uses exact Phase D data
  7. Every data point has explicit date; no bare "recently / today / currently"
  8. Day regime classified via Phase E table; markers cited
  9. Macro regime classified; transition flagged if shifted
  10. Cross-asset coherence classified (coherent-on/coherent-off/divergent)
  11. Thesis Status Board complete (all 5 theses; binary statuses)
  12. =1 contradictory/risk item surfaced if any position moves >2%

  13. Evidence grades on material claims [Grade A-F | source | date]
  14. Geopolitical items passed CFA 3-stage filter (or correctly excluded)
  15. Forward-looking uses % confidence, not HIGH/MEDIUM/LOW
  16. Empty sections skipped entirely; no "nothing to report" padding
  17. Intelligence Gaps flagged if any Phase D error OR >1d stale data
  18. Warning Problems emitted if any divergent / transition / thesis-shift / script-error triggered
  19. Pre-Earnings Monitor for all positions <=10 trading days (script signals)
  20. P&L NOT reported for any no_pnl-flagged position (cost basis not established)
  21. Cross-account holdings report combined exposure (the roster is read from the broker at run time)
  22. Total briefing body <=650 words (3-minute rule); --quick mode <=200 words 22a. GATE-T adjudication (Phase H): every thesis-status change resting on a manual/evidence trigger carries a same-day GATE-T sheet (checked via gate-eval.py, cited in the Evidence cell). No such change -> N/A; --quick -> N/A.

Evidence-grade confidence cap (separate precondition on frontmatter confidence:):

  • 30% of inline grades are Grade C -> cap 70%

  • Any Grade-D in body -> cap 60%
  • 2 STALE data points (>7d) driving decision -> cap 65%

  • Script fallback mode active -> cap 60%
  • Price provenance (C5 mirror of /invest Pre-Output Gate 6a): if the quote feeding any dollar-impact line was non-broker (price-fetcher broker_authoritative: false) during market_session == regular -> cap 60% + flag "price-unconfirmed (non-broker)" in the briefing header. N/A outside regular session; applies to every harness with a regular-session quote.

If frontmatter confidence: exceeds applicable cap: HALT; the owner must authorize override explicitly.

ASCII pre-write scan (Pattern 22): apply Part III sec 3.4 replacement table to all NEW content; byte-scan; HALT on any byte > 127 (modulo pre-existing legacy in unmodified body sections).

FOLLOWUPS:skills sanity: each emitted skill has concrete one-line rationale tied to briefing body; no skill without body evidence.

Priced-in calls schema: each scenario probability + trigger present in frontmatter structured list (input for tomorrow's Brier scoring).

Phase O: Compose briefing + meta.json sidecar

O.1 -- Read ref-output-template.md (sibling of this file) for canonical frontmatter schema, body section order, and meta.json sidecar spec. Do not emit until ref loaded.

O.2 -- Write briefing to Calendar/decisions/briefings/briefing-<YYYY-MM-DD>[-HHMM].md.

Canonical frontmatter (full schema):

---
categories: [decisions]
type: briefing
date: YYYY-MM-DD
regime: <day classification>
macro_regime: <cycle phase>
macro_regime_transition: <true|false>
cross_asset_coherence: <coherent-on|coherent-off|divergent>
portfolio_health: <A|B|C|D|F>
confidence: <integer 0-100>
confidence_cap_rule: <none|grade-c-30pct|grade-d-any|stale-2plus|script-fallback|price-unconfirmed>
conviction: <low|medium|high>
topology: <dw|sequential>
orchestrator_model: <resolved model id; passive>
thesis_statuses:
  theme-alpha: <HEALTHY|WATCH|STRESSED|INVALIDATED>
  theme-beta: <...>
  theme-gamma: <...>
  theme-delta: <...>
  theme-epsilon: <...>
priced_in_calls:
  - event: <name>
    date: YYYY-MM-DD
    bull_prob: <int>
    base_prob: <int>
    bear_prob: <int>
    trigger_level: <number with unit>
prior_brier_score_30d: <float or null; SOURCED FROM brier-ledger.json; n>=3 provisional, n>=5 full>
positions_at_action_zones: [<ticker>@<price>, ...]
followup_skills_count: <int>
created: YYYY-MM-DD
updated: YYYY-MM-DD
status: complete
tags: [topic/market-brief]
related: hot, <materially-discussed-entities-3-7>, <thesis-slugs-invoked>
---

Body section order (skip empty sections entirely; conform to ref-output-template.md updated for v2):

  1. H1 title + Generated stamp + flags (Weekend|Holiday|Pre-Market|Intraday) + BROKERAGE_WARN if applicable
  2. BLUF / Counter / Alternative
  3. Day Regime / Macro Regime / Cross-Asset Coherence / Portfolio Health / Priced In
  4. Markets one-liner
  5. Signal Dashboard
  6. Thesis Status Board
  7. Overnight Earnings (conditional)
  8. Portfolio Movers (What/SoWhat/NowWhat; copy-paste order blocks if action)
  9. Pre-Earnings Monitor (conditional)
  10. Geopolitical (conditional; CFA-filtered)
  11. Alerts (FLASH/PRIORITY per ref-monitoring-rules.md)
  12. Calendar (7 days)
  13. Scenario Bar (conditional; highest-conviction only)
  14. Intelligence Gaps
  15. Warning Problems
  16. (Reserved: a personal section not included in this copy)
  17. Vault
  18. Optionality Map
  19. Today's Focus (1-3 with time-to-decision markers)
  20. FOLLOWUPS:skills block
  21. Footer: Prior-day calibration line (one line from continuity audit) if Brier scoring active

O.3 -- Write meta.json sidecar at Calendar/decisions/briefings/briefing-<YYYY-MM-DD>[-HHMM]-meta.json:

{
  "schema_version": 2,
  "briefing_date": "YYYY-MM-DD",
  "generated_at": "YYYY-MM-DDTHH:MM:SS",
  "mode": "<normal|quick|refresh|...>",
  "topology": "<dw|sequential>",
  "orchestrator_model": "<resolved model id; passive>",
  "net_liquidity": {"walcl": null, "rrp": null, "tga": null, "composite_t": null, "delta_4wk_b": null},
  "insider_overlay": [<0-3 rendered INSIDER lines as objects; [] when empty/skipped>],
  "alerts_fired": [<newly-observed machine-trigger fires this run: {id, trigger, thesis, class, detail, push}; [] when none -- X75 pairs each id against .claude/state/alerts-delivered.log>],
  "priced_in_outcomes": [<resolved ledger calls this run: {event, outcome, brier_contrib}>],
  "script_output_hash": "<sha256 of fetch-prices.py stdout>",
  "indices_snapshot": {<indices block from script>},
  "equities_snapshot": [<equities block>],
  "crypto_snapshot": [<crypto block>],
  "web_searches": [{"query": "...", "url": "...", "grade": "B"}],
  "regime_markers": {"vix": 24.3, "term": "contango", "oas_bps": 412, "classification": "risk-off"},
  "macro_regime_markers": {"yield_curve_10y_3m_bps": -8, "unemployment_trend": "stable_low", "credit_spreads": "tightening", "classification": "late-cycle"},
  "thesis_statuses": {"theme-alpha": "HEALTHY", "...": "..."},
  "priced_in_calls": [<structured list; tomorrow's Brier scoring input>],
  "confidence": 75,
  "confidence_cap_rule": "none",
  "followup_skills": [<structured list>],
  "phases_timing_ms": {"A": 12, "B": 3, "...": 0},
  "pre_output_gate_results": {"item_1_bluf_judgment": "PASS", "...": "PASS"}
}

Sidecar enables machine-parseable retrospective scoring + calibration feedback + audit by future /retro runs. Always written alongside briefing.

O.4 -- ASCII pre-write pass on briefing body + sidecar; HALT on any byte >127 in NEW content.

Phase P: Peripheral atomic updates + Commit + F11 clear + Audit Report

Phase P.0: DELEGATED pre-commit dispatch to vault-classifier-sweep (PREFERRED gate; Phase C wiring 2026-05-02)

MANDATORY (per Execution Rules; SKIPPED in --quick mode per documented exception): dispatch first, no pre-emptive skip outside --quick. Harness-conditional: when the harness exposes no Agent tool, run python tools/vault-audit.py --json inline as the NORMAL path against the identical score/gate thresholds (reported mode: inline (no Agent tool), not a DEVIATION).

Use the Agent tool with subagent_type vault-classifier-sweep. Pass input: {scope: "all"}. Expected return: JSON with 9 classifier keys + totals + score + score_floor_check. If score_floor_check: BREACH (<90) OR gate >0: HALT before atomic apply; surface findings; user must repair before /brief peripheral writes can proceed.

Validate subagent return per its contract; on failure fall through to direct python tools/vault-audit.py invocation as fallback gate. Surface dispatch outcome in Phase P audit.

Phase P.1: Peripheral atomic updates (continues unchanged below)

All four non-briefing updates MUST succeed. Mid-batch failure -> Pattern 13 F.halt: abort remaining writes, F11 stays on, structured report (succeeded / failed / not-attempted), the owner decides (rollback via git checkout -- <paths> or fix-and-retry with idempotency).

Update 1: Calendar/daily/<today>.md ## Market Pulse section

  • Re-Read AT THIS MOMENT (UserPromptSubmit hook may have appended ## Log entries since Phase C)
  • Compute daily_before_sha256 (from re-read body)
  • Replace entire ## Market Pulse section body with: **BLUF:** <BLUF>. **Regime:** <day>/<macro>. **Coherence:** <class>. **Health:** <grade>. **Confidence:** <N>%. Full briefing: [[briefing-<date>[-HHMM]]]
  • Compute daily_after_sha256
  • Assert: body outside ## Market Pulse section byte-exact (compare non-section segments)

Update 2: wiki/hot.md

  • Read; hot_before_sha256
  • Bump last_briefing: to today's ISO
  • Merge pending items additively: any FOLLOWUPS:skills entries with concrete triggers not already present in Pending Items get appended (no removals; Pattern 12 idempotency aligned with /retro spec)
  • Bump updated: to today
  • Compute hot_after_sha256; assert diff is strictly-additive outside frontmatter field bumps

Update 3: Calendar/decisions/sessions-log.md

Append structured entry (canonical schema):

### YYYY-MM-DD[-HHMM] -- /brief
Domain: investing
Day regime: <class>
Macro regime: <phase>
Portfolio Health: <grade>
Confidence: <N>%
BLUF: <one-line>
Thesis shifts: <list of HEALTHY->WATCH transitions OR "none">
Followup skills: <count>
Artifact: [[briefing-<date>[-HHMM]]]

Update 4: wiki/entities/tickers/<T>.md for each entity in briefing related:

  • SCOPE: only entities in briefing related: field (3-7 typical), NOT every ticker mentioned in body. Keeps fan-out bounded; consistent with /invest analysis-to-entity back-link discipline.
  • For each entity:
    • Read; entity_before_sha256
    • Append under Recent section (create if absent): - [YYYY-MM-DD] Mentioned in [[briefing-<date>[-HHMM]]] -- <one-phrase context>
    • Bump updated: to today
    • Compute entity_after_sha256; assert strictly-additive diff

Update 5: Calendar/decisions/briefings/brier-ledger.json (machine-owned; brief-vnext 2026-07-04)

  • Schema: {calls: [{event, date, bull_prob, base_prob, bear_prob, trigger_level, source_briefing, outcome: null|bull|base|bear, resolved_date, brier_contrib}], rolling_30d: float|null}. Create file on first run.
  • APPEND today's new priced_in_calls entries (source_briefing = today's briefing stem; outcome null).
  • RESOLVE any open call whose event date has passed: score outcome from Phase D actuals (CPI prints, FOMC, earnings reactions); resolution is a one-line judgment with the evidence cited in the sidecar; set resolved_date + brier_contrib; recompute rolling_30d.
  • Machine-owned JSON (precedent: meta.json sidecars already live under Calendar/); body-preservation N/A; stdlib-json parse gate before write.
brief(daily): YYYY-MM-DD[-HHMM] -- <regime> day, <macro_regime> macro, <health> health

Briefing at Calendar/decisions/briefings/briefing-<date>[-HHMM].md
Meta sidecar at briefing-<date>[-HHMM]-meta.json

BLUF: <one-line>
Day regime: <class> (VIX <X> [cont|bkwd], OAS <Xbps>, ...)
Macro regime: <phase> (<curve, unemp, credit>)
Cross-asset coherence: <class>
Portfolio Health: <grade>
Confidence: <N>% (cap: <rule or none>)
Conviction: <level>
Thesis statuses: theme-alpha=<X>, theme-beta=<X>, theme-gamma=<X>, theme-delta=<X>, theme-epsilon=<X>
Followup skills: <count>
Prior 30d Brier score: <float or n/a>

Peripheral atomic updates:
 - Calendar/daily/<today>.md  (Market Pulse summary + linkback)
 - wiki/hot.md  (last_briefing bump; additive pending merge)
 - Calendar/decisions/sessions-log.md  (structured entry)
 - wiki/entities/tickers/*.md  (<N> symmetric back-links)

F14 narrow stage each file explicitly. Never git add -A / git add ..

Post-commit F17 verify: git log -1 --format='%B' HEAD | grep -c '^Co-Authored-By:' must equal 0. If found, HALT; ask the owner (do NOT silently proceed).

Clear F11: rm .claude/state/auto-commit-disabled. Assert absence.

Audit Report (stdout to the owner):

## /brief v2 run complete

Briefing:           briefing-<date>[-HHMM].md
Meta sidecar:       briefing-<date>[-HHMM]-meta.json
Day regime:         <class> (markers: VIX=X term=X OAS=X)
Macro regime:       <phase> (curve=X unemp=X credit=X)
Cross-asset:        <coherent-on|coherent-off|divergent>
Portfolio Health:   <grade> (<justification>)
Confidence:         <N>% (cap rule: <none|grade-c-30pct|grade-d-any|stale-2plus|script-fallback>)
Conviction:         <level>
Thesis statuses:    theme-alpha=<X>, theme-beta=<X>, theme-gamma=<X>, theme-delta=<X>, theme-epsilon=<X>
Thesis shifts:      <list or "none">

Files written:      5 + 1 sidecar + <N> entity back-links = <total>
Entity back-links:  <list of tickers>
Followup skills:    <count> (<skill names>)

Script status:      <ok|fallback|errors on: X,Y>
Web searches:       <count>
Pre-Output gate:    22/22 PASS
F11:                cleared
F17:                Co-Authored-By absent
Working tree:       clean

Prior-day calibration: <one-line from continuity audit, or "n/a (insufficient prior calls)">

Pattern 21 / Standing Rule 16 -- companion refs (created)

The following ref docs EXIST on disk as SIBLINGS of this file, as companion reference material (each 60-77KB, built via Deep Research per Standing Rule 16). Consult them when the inline procedural depth is insufficient for a deep-dive:

  • ref-regime-taxonomy.md -- full day + macro regime taxonomy, transition matrices, historical base rates
  • ref-evidence-hierarchy.md -- Grade A-F evidence rubric, freshness tiers, contradiction tiering
  • ref-briefing-structure.md -- PDB ICD 203 + hedge fund morning note structural taxonomy

SKILL.md v2 covers their content inline at procedural depth: 6-regime table in Phase E, 4-class macro table also Phase E, Grade A-F rubric in Quality Rules (formerly Quality Standards), briefing structure in ref-output-template. Keep procedural depth inline as the operative path; the companion refs carry the extended taxonomy for deep-dives.

Failure Taxonomy -- see ref-brief-operations.md Section 2. HALT semantics (F.halt, F11-stays-on, F17) are normative there.

Coordination -- see ref-brief-operations.md Section 3 (shared infra + division-of-concerns matrix + consumers)

Examples -- see ref-brief-operations.md Section 1 (8 worked mode examples)

Ref docs (all siblings of this file unless pathed)

  • ref-output-template.md (v2; consumed in Phase O for canonical frontmatter + body section order + meta.json sidecar spec + daily-note linkback pattern)
  • ref-macro-data-sources.md (ADD-NOW 2026-06-05; consumed in Phase E.0 for the FRED series catalog incl. net-liquidity composite + day/macro regime mappings + availability-guard pattern + C6/C8 caveats)
  • ref-brief-operations.md (brief-vnext Diff 1 extraction 2026-07-04; Section 1 worked examples, Section 2 failure taxonomy [HALT semantics normative], Section 3 coordination matrix + consumers)
  • .claude/workflows/brief-research.js (TOPOLOGY=dw Wave-1 acquirer bundle; invoked per Phase A.8)

Companion reference docs (Pattern 21; siblings of this file -- extended taxonomy consulted when inline procedural depth is insufficient):

  • ref-regime-taxonomy.md (day + macro regime taxonomy, transition matrices, base rates)
  • ref-evidence-hierarchy.md (Grade A-F rubric, freshness tiers, contradiction tiering)
  • ref-briefing-structure.md (PDB ICD 203 + morning-note structural taxonomy)

Related skills

  • /networth -- portfolio snapshot complement (Phase 15 portfolio refresh from /brief v1 retired here)
  • /invest <ticker> -- per-ticker analysis triggered by FOLLOWUPS
  • /challenge thesis-<slug> -- thesis stress-test triggered by status shift in Phase H
  • /decide -- structured decision record if action proposed in Phase J Now What
  • /retro -- session-end retrospective consumes /brief sessions-log entry

Phase O.0 -- Pre-commit /vault audit gate (v2.0; CAT-3 prevention-architecture parity)

After composing all target file modifications IN MEMORY but BEFORE atomic write:

  1. Write each composed file to a tmp dir under wiki/research/test-tmp/.precheck/brief-<slug>/
  2. Run python tools/skill-precheck.py <tmp-files...> --skill /brief
  3. Parse exit code: 0 -> proceed; 2 -> HALT with diagnostic
  4. Body-scope wikilink validation: per /retro v2.2 Phase D pattern, scan composed body text for unresolved [[<target>]] and mechanically de-link unresolved targets (vault-resolved keep / MEMORY_PREFIXES rewrite as [[memory:<stem>]] / placeholder leave / else strip). Fence-aware (skip ``` fenced + inline code).
  5. Bypass: CLAUDE_VAULT_BYPASS_VALIDATOR=1 (logged to .claude/state/bypasses-<date>.log)

Defense-in-depth on top of PreToolUse pre-write-validator.py + PostToolUse wikilink-check.py / frontmatter-check.py / orphan-check.py. The Phase O.0 gate prevents broken composition from reaching disk in the first place.

/brief-specific risk: /brief writes to ~5 append-mode files daily (sessions-log + hot.md + daily + 3-7 entity back-links) -- highest write-fanout among HIGH-risk skills.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.