agentleFS
KemetraKemetra/Seshat-BI/.claude/skills/portfolio-watch/SKILL.md

portfolio-watch

Skill2 starsChanged 4 months agoApache-2.0View on GitHub

Show the recurring, read-only portfolio summary for the Seshat BI repo -- every governed scope's covered-dimension findings (source drift, contract/metric drift, dashboard-intent divergence, readiness, approvals, review), open blockers, human-attention flag, and ONE prioritized next action, plus a baseline diff against the last run (new/resolved/unchanged). Use when someone asks "run portfolio watch", "what changed since last week", "show me the recurring summary", or "which scopes need a human, and what changed". READ-ONLY and invoke-and-present only: it AGGREGATES evidence that shipped surfaces already produce (it runs NO new check, re-derives NOTHING); it persists a LOCAL baseline snapshot under `.seshat/watch/` so the next run can diff; it introduces NO new gate/rule/approval mechanism and emits NO fabricated health/confidence/priority score (every line is a shipped categorical status, a shipped finding enum, or a measured magnitude traced to a committed source).

3 other projects use a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

6 rules. 0 are common, 1 is distinctive, and 5 are very specific to this project.

  • 0

    Common

    Most of the 3 edited versions we read keep these.

  • 1

    Distinctive

    This author's own choices. Worth a look.

  • 5

    Very specific

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

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

    Aggregates, never re-derives: Every finding is sourced from a shipped surface's already-committed output (drift.py/drift_semantics.py, …

    line 31
  2. Very specific:

    No score, ever: Only the four readiness spine statuses, the shipped categorical finding enums (relayed verbatim), and measured magnitudes …

    line 40
  3. Very specific:

    Every dimension is one of the five closed states: covered (cites its evidence) | [PENDING LIVE] | stale (cites captured-at vs current HEAD) …

    line 67
  4. Very specific:

    No prior snapshot (first run, or a corrupt one) -> every condition is current_condition_no_baseline -- explicitly NOT new -- and the …

    line 87
  5. Distinctive:

    If asked for "a portfolio health score", "a percent-ready", or "one confidence number", DECLINE: cite hard rule #9 and this skill's …

    line 116
  6. Very specific:

    After a run, git status shows zero modified files under mappings/ (or any other committed per-scope artifact) -- the only new/changed path …

    line 122

“Common” is measured: kept by at least 6 in 10 of the 3 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

portfolio-watch

Portfolio Watch answers one recurring question: since we last looked, what changed, what still needs a human, and what is the one next thing to do per scope? It is the baseline-diffable, multi-dimension EXTENSION of the point-in-time retail-control-room roll-up (F012) -- same read-only, aggregate-never-re-derive posture, plus a persisted snapshot and the drift/semantic-drift/dashboard-intent-divergence dimensions the control room does not fold in.

Scope wall (read first)

  • Aggregates, never re-derives. Every finding is sourced from a shipped surface's already-committed output (drift.py/drift_semantics.py, metric_drift.py, semantic_audit.py/report_intent.py, readiness_projection.py/readiness_classify.py, approval_inbox.py, review_integration.py/review_pack_export.py). It runs NO new check and calls NONE of those surfaces' own comparators itself -- it reads what they already recorded and cites the path.
  • No new gate / rule / approval mechanism. This is a Tier-5 read/summarize/derive companion (roadmap binding rule), not another governance engine. seshat check's rule inventory and exit behavior are unchanged by this skill's existence.
  • No score, ever. Only the four readiness spine statuses, the shipped categorical finding enums (relayed verbatim), and measured magnitudes (counts/rates/deltas) traced to a committed source appear anywhere. A health/confidence/priority/quality number is never emitted (hard rule #9) -- if asked for one, decline and point here.
  • No live database in the MVP. The governed-scope set comes from committed mappings/*/readiness-status.yaml paths (the same set status_surface / readiness_projection already track), never a live metadata query. A dimension whose evidence needs a live re-profile with no DSN configured is truthfully [PENDING LIVE].
  • Relays Principle-V, decides none of it. A grain/PII/returns/rollup/identity/approval condition is relayed with a named owner; this skill originates no ruling on any of them. requires_human_attention fires independently of category rank whenever an unmet approval or a relayed Principle-V/PII drift blocker is present -- it can never be buried by a higher-ranked, less dangerous condition.
  • "Recurring" = re-runnable + baseline-diffable, not a scheduler. There is no cron, no hosted monitoring, no continuous-run surface here or planned by this skill.

What it aggregates (the six covered dimensions)

DimensionSource it relaysTruthful degradation
Source driftdrift.py / drift_semantics.py committed findingsno live re-profile recorded -> [PENDING LIVE]
Contract/metric driftmetric_drift.py verdicts (a committed record of a retail semantic-check run)no recorded run -> not_applicable_with_reason
Dashboard-intent divergencesemantic_audit.py / report_intent.py committed findingsno recorded run -> not_applicable_with_reason
Readiness (+ changed readiness)readiness_projection.py / readiness_classify.pya malformed readiness-status.yaml -> unreadable for that scope, never dropped
Stale/missing approvalsapproval_inbox.py(read directly; always covered or unreadable)
Review (tables requiring attention)review_integration.py / review_pack_export.py committed resultno recorded result -> not_applicable_with_reason

Every dimension is one of the five closed states: covered (cites its evidence) | [PENDING LIVE] | stale (cites captured-at vs current HEAD) | not_applicable_with_reason (names why) | unreadable (names the unknown schema version / read error). No state is ever silently upgraded to covered, and one scope's read error never aborts the run -- a partial portfolio is summarized for the covered scopes, with the un-evidenced ones listed in portfolio.scopes_with_no_evidence.

The one prioritized next action per scope

Selected by the SHIPPED fixed readiness_classify rank (approval > grain > live_validation > artifact > readiness) over that scope's own open readiness blockers; the action text is that scope's own recorded next_action, RELAYED verbatim -- never synthesized. Two scopes tying on the same top category each surface their own action; the rank is a committed lookup, never broken by an invented number.

The baseline diff (the genuinely new part over the control room)

Each run reads the prior local snapshot (.seshat/watch/snapshot.json), diffs the current magnitude-free Condition Keys against it, and writes a fresh snapshot:

  • No prior snapshot (first run, or a corrupt one) -> every condition is current_condition_no_baseline -- explicitly NOT new -- and the summary states that no baseline was available (mirrors drift.py's observed=None first-run honesty).
  • A prior snapshot exists -> each condition is new / resolved / unchanged. A magnitude wiggle on the SAME class/locator never churns as new/resolved (duplicate suppression) -- only a genuinely new or genuinely resolved condition moves.
  • A scope added/removed between runs is reported as a scope-level change, never misattributed as a condition change inside a scope that no longer exists (or does not exist yet).

Run it

Invoke build_portfolio_watch_summary / run_portfolio_watch from seshat.portfolio_watch (the pure aggregator library), or the one narrow CLI surface for CI/agent-less use:

retail watch --format json
retail watch --format text

Present the result: per scope, current_stage, the covered-dimension findings (state + class + evidence + owner where relevant), open_blockers, requires_human_attention (+ owner when set), the one prioritized_next_action, and the change_labels against the prior snapshot. Close with the portfolio block (measured scope_count / scopes_requiring_attention_count / scopes_with_no_evidence) and note that the run wrote a fresh local snapshot for next time.

No fake confidence (the guardrail)

If asked for "a portfolio health score", "a percent-ready", or "one confidence number", DECLINE: cite hard rule #9 and this skill's frontmatter, and return the four statuses + the shipped categorical enums + the measured magnitudes with their source paths instead.

Read-only proof

After a run, git status shows zero modified files under mappings/ (or any other committed per-scope artifact) -- the only new/changed path is the local .seshat/watch/snapshot.json (git-ignored by default). No approval is recorded, no readiness stage moves to pass, no database is opened, nothing is published.

See also

  • The closest shipped sibling (same posture, no baseline/no extra dimensions): .claude/skills/retail-control-room/SKILL.md (F012). Portfolio Watch EXTENDS this posture; it does not fork or replace it.
  • The aggregator library + snapshot format: src/seshat/portfolio_watch.py.
  • The tool doc: docs/tools/portfolio-watch.md.
  • The feature spec: specs/131-portfolio-watch/spec.md (+ data-model.md, contracts/).
  • The gates it reads (never re-runs): retail-govern / seshat check, retail-semantic-check, retail-validate; the readiness model: docs/readiness/readiness-model.md; the source-drift precedent: docs/readiness/source-drift.md.
  • The ratified CLI-vs-skill decision behind the one narrow retail watch surface: docs/roadmap/decisions/cli-verbs-vs-skill-driven.md.

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.