market
Skill0 starsChanged 54 days agoMITView on GitHub
Market mega-skill: surveillance, screening, signals, regime, and earnings behind one dispatcher. /market watch = severity-ranked alert board (triggers/earnings/macro/filings); /market screen `<mode>` = quant screener over the local factor store; /market signals = full signal-library dashboard; /market regime = current regime assessment; /market earnings = binary-events calendar. Read-only; cites every number; never recommends trades.
- Reads credentials
14 other projects use a version of this.
Adopted by
Most recently taken up by
- AlejandroGodius/claude-skillstheir own version, first seen Nov 16, 2026
- nazmirket/skillstheir own version, first seen Nov 15, 2026
- AISA-skills/financial-data-skillstheir own version, first seen Nov 1, 2026
- jtmuller5/small-business-skillstheir own version, first seen Oct 20, 2026
- ollie88skwda/claude-skillstheir own version, first seen Oct 13, 2026
- Antthonyyy/marketing-agents-kittheir own version, first seen Oct 13, 2026
Used with
Projects with it also use market-ads, market-audit and 10 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.
AlejandroGodiusSkillmarket-ads
Also in 4 projects that use this
AlejandroGodiusSkillmarket-audit
Also in 4 projects that use this
- +10 more from AlejandroGodius
Other versions14 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 14
- zubair-trabzada/ai-marketing-claude★ 2.6ksame name
- rediumvex/ai-marketing-claude★ 38same name
- khalidgraphy/claude-market-skills-by-KM★ 10same name
- AISA-skills/financial-data-skills★ 4same name
- Yafan-Yang/ai-trading★ 4same name
- tal7aouy/marketkit★ 2same name
- AlejandroGodius/claude-skills★ 0same name
- Antthonyyy/marketing-agents-kit★ 0same name
- baofeng-tech/agent-skills-so★ 0same name
- hmzainjamil/ai-marketing-claude★ 0same name
4 more not listed
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository18 other files it gives its agents.
The file
We split this skill into the rules it gives an agent
10 rules. 0 are common, 7 are distinctive, and 3 are very specific to this project.
0
Common
Most of the 12 edited versions we read keep these.
7
Distinctive
This author's own choices. Worth a look.
3
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 93
NO ADVICE LEAK: forbidden "buy", "short", "trim", "you should sell", "buy puts", "consider exiting", sizing or timing language.
- Distinctive: line 98
NO STALE LITERALS: parse prices, levels, holdings at run time.
- Distinctive: line 104
No network calls; use the store/refs plus already available current-session broker read evidence for optional HELD status.
- Very specific: line 106
Never write/edit/delete any file unless a mode explicitly allows it (SCREEN --save).
- Distinctive: line 139
READ IT before running the macro lane the first time.
- Distinctive: line 162
Never recommend filling, canceling, or moving an order.
- Distinctive: line 169
Always show: date + TENT marker, ticker, exp.
- Distinctive: line 272
Every result row carries: ticker, score components, key metrics, latest analysis rating with source file + date, HELD yes/no.
- Very specific: line 319
Only ~82 tickers have XBRL JSONs, only ~30 extended (FCF/share count): value/quality/oversold universes are far smaller than 107.
- Very specific: line 382
READ-ONLY, stdout only. Sources: wiki/investing/benchmarks/macro-regime-tables.md (Fed plumbing + curve + credit + yen + VIX percentile.
“Common” is measured: kept by at least 6 in 10 of the 12 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.
Files
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.
/market -- consolidated market desk (watch + screen + signals + regime)
Created 2026-08-24 by consolidation: absorbs /watch, /screen, plus the
signals and regime surfaces (no standalone /signals or /regime skill files
existed; their logic lives in the signal library and macro-regime refs this
skill reads). Originals archived at
.agents/skills/_archive/consolidated-2026-08-24/ (see INDEX.md); the ref
docs moved alongside into this directory (ref-watch-logic.md,
ref-screening-logic.md). All rules carried over verbatim in substance from
the source skills; each subcommand routes to its logic.
This skill screens, measures, classifies, and routes only. Trade decisions,
sizing, and doctrine verdicts belong to /invest, /gate f, /portfolio,
and the operator.
Invocation and dispatch (Pattern 6 -- deterministic, from $ARGS only)
| $ARGS | Route |
watch or watch dashboard | MODE WATCH, full board (all lanes) |
watch triggers | MODE WATCH, entity-surveillance triggers lane only |
watch earnings | MODE WATCH, upcoming prints for HELD names only |
watch macro | MODE WATCH, DGS10 / DFII10 / VIX / liquidity vs gates |
screen <mode> [--top N] [--save] | MODE SCREEN (modes below) |
signals | MODE SIGNALS |
regime | MODE REGIME |
earnings | MODE WATCH, earnings lane rendered calendar-style |
Unknown mode -> print usage table and stop. No args -> usage table, run
nothing. READ-ONLY + OFFLINE by default: writes nothing anywhere except
MODE SCREEN with explicit --save. Never fetches live quotes or filings.
Forbidden reads: .raw/, private/, finance/, credentials/,
.env*, auth.json, *.local.md. ASCII only.
Shared inputs (all local; verify freshness before trusting)
Efforts/osanwe-v2-overhaul/_work/factors.db(READ-ONLY sqlite open,file:...?mode=ro) -- bars (ticker,date,close) + macro factors (ticker='MACRO': DGS10, DFII10, VIXCLS always; STLFSI4, WRESBAL, WALCL, RRPONTSYD, SOFR, FEDFUNDS, DEXJPUS for the macro lane). Report MAX(bar date) used.wiki/maintenance/surveillance/entity-triggers-*.md-- LATEST dated file; generated bytools/entity-surveillance.py --proximity 8. If older than 3 trading days vs the store's last bar date: one INFO staleness banner with the rerun command; never regenerate silently.wiki/research/ref-earnings-calendar.md-- HELD roster + dated print table (TENTATIVE vs CONFIRMED, Binary flags, key metrics).wiki/research/ref-fed-liquidity-engine.md-- liquidity engine verdict + tripwire thresholds/action biases.wiki/investing/filings/<T>/per-ticker filing lists (form,date); 8-K recency screen only, no document fetches; XBRL JSONs for fundamentals._work/signals-library-data.json-- r1m..r12m, MA states, RSI14, vol_ratio, beta60 convenience snapshot; recompute from bars when stale.Atlas/sources/investing/ref-quantitative-signals-library.md+ generatedwiki/research/ref-composite-scoring.md-- signal definitions + ranked tables (cite generation dates).wiki/investing/benchmarks/macro-regime-tables.md-- Fed policy plumbing (WALCL/WRESBAL/SOFR/RRP + curve + credit + yen + VIX percentile).- Current-session broker read evidence, when already available, supplies HELD status. Otherwise mark holdings UNVERIFIED; saved history cannot establish current membership. This offline skill does not initiate broker calls.
- Wave analyses (
_work/wave*-analysis.md) + analysis-file globs for ratings/zones; NEVER Atlas/concepts watchlist rows (documented stale prior incident).
Quality Rules (global execution rules; bind EVERY mode)
- Cite every number:
(computed: <formula id>, factors.db bars thru <date>),(per <path>, filed <date>), or(per <doc>, as-of <date>). Bare figures are defects. - NO ADVICE LEAK: forbidden "buy", "short", "trim", "you should sell", "buy puts", "consider exiting", sizing or timing language. Allowed: measured statements, classifications ("mean-reversion watch", "avoid / risk-off candidate"), verbatim attributed doctrine quotes, pre-committed responses routed to /gate f or /decide.
- NO STALE LITERALS: parse prices, levels, holdings at run time; nothing from memory or skill prose.
- BLIND SPOTS ARE FINDINGS: missing files, tentative dates inside 5 sessions, store staleness, skipped names -- each gets its own line.
- REGULAR-CLOSE ANCHORING: all doctrine evaluations use regular-session closes; AH/pre-market moves are context, never triggers.
- No network calls; use the store/refs plus already available current-session broker read evidence for optional HELD status. Missing holdings stay UNVERIFIED.
- Never write/edit/delete any file unless a mode explicitly allows it (SCREEN --save); regeneration commands are surfaced, run only on explicit operator request.
- The model never judges sizing, doctrine bands, or verdicts -- those belong to /gate (tools/gate-eval.py), /portfolio advise, kernel doctrine.
MODE WATCH (= former /watch [dashboard|triggers|earnings|macro])
Scan the vault's standing data for anything moved close enough to a pre-committed decision point to demand attention; print a dated, severity- ranked alert list. Every alert carries numbers, source citation, and the response doctrine ALREADY commits to -- the operator decides whether to act, never what the reaction would be.
Phase W-A load context
Route on args. Resolve inputs 1-5 above; record each file's date for the freshness line; missing input -> gap row (R-STALE), never an estimate. Triggers lane: parse latest surveillance table into (ticker, last, trigger-class, level, distance); classify HELD vs watchlist off the calendar roster. Earnings lane: parse print rows (drop rolling index rows); compute trading-days-until off the session grid. Macro lane: pull latest DGS10 / DFII10 / VIXCLS (+ tripwire series when macro mode or any CRITICAL fires) with as-of dates. Filings lane: newest 8-K per covered ticker.
Phase W-B evaluate and classify
Apply ref-watch-logic.md Section 3 evaluations to the macro series
(rate_shock window math computed fresh from the store -- no hardcoded
readings). Grade every candidate through the master severity table (ref
section 2); collapse duplicate-class trigger neighbors per ref section 5.
READ IT before running the macro lane the first time.
Phase W-C render
Emit exactly one Output-format block matching the mode; sort CRITICAL -> WARNING -> INFO, nearest distance / soonest date within tier; attach the catalog response id + one-line pre-committed response to every alert. Quote doctrine thresholds and engine action biases verbatim with source; never paraphrase tighter or looser.
Phase W-D close out
Print lane summaries + data freshness line. If nothing fired anywhere say exactly "nothing material to report" (quiet-day rule). Nothing written to disk.
Lane logic (thresholds summarized; normative detail in ref-watch-logic.md)
Triggers lane:
- kill-stop touched/crossed on HELD name -> CRITICAL
- kill-stop within 2 pct (held), any held trigger within 2 pct -> WARNING
- LIMIT buy within 1 pct -> WARNING (fill imminent; order stands per doctrine)
- MA-anchor / zone / other trigger within 8 pct -> INFO (WARNING if held) Never recommend filling, canceling, or moving an order; route to /gate f.
Earnings lane:
- CONFIRMED/TENTATIVE print within 2 trading days, BINARY flag, no pre-print plan on record -> CRITICAL
- within 5 trading days -> WARNING (earnings-week protocol engaged)
- within 10 trading days -> INFO (heightened monitoring) Always show: date + TENT marker, ticker, exp. move, binary flag, key metrics.
Macro lane (deployment gates + liquidity tripwires):
- DFII10 rate_shock (>= +75 bp over 60 published sessions, nominal confirm DGS10 >= +25 bp, entered after 3 consecutive fire sessions; exit < 60 bp) ACTIVE -> CRITICAL (0.5x deployment throttle).
- VIXCLS > 35 all last 5 observations -> CRITICAL (regime halt 0.0x); VIXCLS > 22 all last 5 -> WARNING (throttle 0.5x); any of last 5 > 30 -> WARNING (reserve-release vix-spike condition armed).
- DGS10: DISCLOSURE ONLY -- level gate retired 2026-07-30; report level + delta, never treat a level as a deployment blocker.
- Liquidity tripwires (engine Section 8): DEXJPUS carry-unwind signature -> CRITICAL; WRESBAL < 2,800 B (4-wk avg) -> CRITICAL; SOFR-FF > +10 bp 3+ sessions off quarter-end, RRPONTSYD re-inflating > 100 B, STLFSI4 crossing 0 / +0.33, WALCL renewed weekly declines > 30 B x3 -> WARNING. Quote the engine's Section 7 VERDICT line verbatim beneath the scoreboard.
Filings lane: newest 8-K within 5 calendar days on HELD name -> WARNING; on watchlist name -> INFO. Report form + date + doc filename only.
Severity rubric (full table in ref-watch-logic.md): CRITICAL = doctrine demands a decision THIS SESSION (gate/halt active, kill-stop touched on a holding, reserve-release met, carry-unwind signature, imminent unprepared binary print) -- pre-committed response runs through /gate f or /decide, no discretionary overrides. WARNING = act-on-your-plan item; response is documented; no forced action. INFO = situational awareness. Multipliers stack by MIN across active states, never multiplied (doctrine).
Output format (every mode)
== /market watch <mode> -- <YYYY-MM-DD> ==
<one line; or exactly "nothing material to report" per quiet rule>
Data freshness: triggers <file-date> | store last bar <date> |
calendar updated <date> | liquidity engine <date>
CRITICAL:
TICKER -- class -- detail with numbers [citation]
Response: <pre-committed doctrine response, one line>
WARNING: ...
INFO: ...
Lane summaries: triggers N alerts | earnings next: A, B, C | gates: <state>Sort as specified above; empty lane -> "clear". F11-style commit machinery does not apply to this mode (writes nothing).
MODE SCREEN (= former /screen <mode>)
Rank the 107-instrument universe (factor-store bars) with the vault's own
quant layers: composite scoring (ref-composite-scoring), the signal library
(ref + _work/signals-library-data.json), and EDGAR XBRL pulls
(wiki/investing/filings/<T>/). Normative formulas:
ref-screening-logic.md (this directory). A screen hit is a CANDIDATE LIST,
never a verdict; acting requires a full /invest run.
Mode routing
| Invocation | Screen | Ranked by | Hard filters |
screen momentum | Trend leaders | MOM component desc (Method 1) | RSI14 <= 70 AND px > MA50 AND >=200 bars |
screen value | Cheap vs fundamentals | P/S asc (P/E display tiebreak) | TTM FCF > 0 AND rev YoY > 0 AND XBRL present |
screen quality | Fundamental strength | FUND component desc | margin-stable AND dilution <= 3%/4q (share-count data required) |
screen oversold | Mean-reversion candidates | RSI14 asc | RSI14 < 30 AND px < MA200 AND fund_pos >= 2 |
screen breakdown | Technical damage | vol_ratio desc among fresh breaks | broke 20-session low (or MA) <=5 sessions ago AND vol20/vol60 > 1.2 |
screen setup | Stated entry zones | effective R/R desc | wave zone AND R/R >= 3:1 AND zone NOT voided |
Unknown mode -> print the mode table and stop. Default --top is 10.
Screening logic (normative summary; exact math in ref-screening-logic.md)
- momentum -- MOM = 0.30ret21d + 0.25ret63d + 0.25dist_MA50 + 0.20dist_MA200, each clipped to [-50,+50]% then shifted to [0,100].
- value -- TTM revenue and TTM NI = trailing four REPORTED quarters
(fiscal-Q4 derived as FY minus 3 quarters when needed); TTM FCF from
-xbrl-extended.jsonFreeCashFlow_computed (fallback: operating cash flow minus capex concepts). P/S = mktcap / TTM revenue; P/E shown only when TTM NI > 0. EPS-based P/E IMPOSSIBLE in this corpus (zero EPS points). - quality -- FUND component = mean of available percentile-ranked sub-scores (rev YoY growth, TTM net margin, cash/LTD debt; >=2 of 3 required). Stability gate: TTM net-margin stdev trailing 4q <= 5pp AND current TTM margin >= year-ago - 5pp. Dilution gate: ShareIssued latest vs 4 quarters prior <= +3%. No share-count data -> excluded from top-10, listed BLIND SPOTS.
- oversold -- fund_pos counts positive members of {revenue YoY, TTM net margin, TTM FCF}; rank ascending RSI14; show depth below MA200.
- breakdown -- fresh break = pierced running 20-session closing low within last 5 sessions OR crossed below MA50/MA200 within last 5; expansion = vol20/vol60 > 1.2 (report vol_pctile_own too). Label: avoid / risk-off.
- setup -- parse wave bodies for zones/stops/targets/stated rr_at_zone_* lines. Effective R/R = best STATED zone R/R, else computed (target - zone_mid)/(zone_mid - stop). Keep only R/R >= 3.0. VOID CHECK against live bars: close beyond stop/cancel since analysis date -> zone VOID (STOP-BREACHED law) -- drop to voided list, never recompute the bracket. Known rejection reasons travel with the name (e.g. BWXT raw 3.16:1 REJECTED, stop inside noise band).
Every result row carries: ticker, score components, key metrics, latest analysis rating with source file + date, HELD yes/no.
Mode columns: momentum MOM|r21d|r63d|vsMA50|vsMA200|RSI14|COMP; value P/S | P/E | TTM rev | TTM FCF | rev YoY | NI marg TTM | FUND; quality FUND | rev YoY | NI marg TTM | cash/LTD | margin stdev | dilution 4q; oversold RSI14 | vsMA200 | rev YoY | NI marg TTM | TTM FCF | fund_pos; breakdown vol_ratio | vol pctile | off-low | break type | days since break; setup zone | stop | target | R/R mid | R/R best | dist to zone | rating.
Provenance footer (every run): factors.db bars thru <date> |
ref-composite-scoring.md as-of <date> | signals-library-data.json generated
<date> | xbrl filed dates inline | ratings globs | HELD file date; caveats
line for stale-holdings / stale-fund / thin history.
Procedure
Phase B-A parse: open db read-only; MAX(date) = as-of; split SCORED (>=200
bars for MA200 modes; >=64 others) vs SKIPPED with reasons (short-history
IPOs cannot produce valid MA200/RSI readings -- SKIPPED, not zeros); load
HELD set with the 7-day gate. Phase B-B compute: run the mode pipeline from
ref-screening-logic Sections 2-7; apply hard filters; sort; truncate --top;
attach composite components, rating + file/date, HELD flag. Phase B-C
render the table + BLIND SPOTS + provenance footer (format below). Phase
B-D close out: --save writes ONE dated snapshot
wiki/investing/snapshots/screen-<mode>-<YYYY-MM-DD>.md (same-day collision
-> -HHMM; canonical wiki frontmatter; ASCII); otherwise nothing on disk;
print NEXT line pointing at /invest <ticker>.
Output format:
== /market screen <mode> ==
As-of: <MAX bar date> | universe: <scored>/<107> scored, <k> skipped
rank | ticker | <component cols per mode> | rating (src date) | HELD
-----+--------+-----------------------------+------------------+-----
1 | <TICKER> | ... | HOLD (wave1-example) | YES
BLIND SPOTS: <unscored / stale-fund / no-xbrl / voided-zones lines>
NEXT: /invest <ticker> for any name worth acting onScreen pitfalls
- Composite-doc ranks are as-of ITS generation date; recomputing from bars diverges when bars moved since. Cite your as-of; never mix both in one table.
- Only ~82 tickers have XBRL JSONs, only ~30 extended (FCF/share count): value/quality/oversold universes are far smaller than 107; print real denominators.
- EPS structurally unavailable corpus-wide; P/E here means mktcap / TTM NI.
- YoY compares LQ to the same fiscal quarter 330-400 days earlier; naive index arithmetic crosses fiscal calendars.
- Wave
rr_ratiofrontmatter is often AT SPOT and fails 3:1 while the ZONE R/R passes (ETN spot 0.94 vs zone-low 3.33) -- setup keys on ZONE R/R. - A zone whose cancel level traded after publication is DEAD even when its printed R/R looks attractive (VOID CHECK).
- HELD comes only from current-session broker read evidence. Otherwise its status is UNVERIFIED; do not assert unheld status or silently present a held-only earnings/filter result as complete. Continue the requested non-account screening with the coverage gap disclosed.
- Weekly hygiene: after SUN 07:30 calibration ingest refreshes factors.db,
regenerate the signals snapshot (
_work/signals-lib-compute.pythen-render.py) or rely on live-from-db recomputation. - Material extension of this mode: run
/gate bfirst (root AGENTS.md invariant). Formula changes land in ref-screening-logic.md + a dated note, never silent edits.
MODE SIGNALS (= full signal-library dashboard)
Render the complete signal library state for one pass: the dashboard view of what every tracked indicator says right now. READ-ONLY, stdout only.
Sources: Atlas/sources/investing/ref-quantitative-signals-library.md
(definitions + ranked tables; cite generation date) and
_work/signals-library-data.json (snapshot values; if stale vs db MAX bar
date, recompute live from bars using the same formula parity rules as MODE
SCREEN: Wilder RSI(14), simple MA50/MA200, trading-day-offset returns).
Render sections, each citing its source:
== /market signals -- as-of <MAX bar date> ==
MOMENTUM BLOCK: r1m/r3m/r6m/r12m leaders + breadth (% of universe above
MA50 / MA200) [signals data or recomputed]
MEAN-REVERSION BLOCK: most oversold (RSI14 asc) + freshest breakdowns
(vol_ratio) -- classification labels only, no trade language
VOLATILITY/BETA BLOCK: beta60 extremes, vol_ratio outliers, VIXCLS level
+ percentile [factors.db MACRO]
MACRO FACTORS BLOCK: DGS10/DFII10 levels + deltas, liquidity series state
(STLFSI4/WRESBAL/WALCL/RRPONTSYD/SOFR/FEDFUNDS/DEXJPUS) with as-of dates
COMPOSITE OVERLAY: top/bottom COMPOSITE names per ref-composite-scoring.md
AS-OF ITS generation date (never mixed with recomputed ranks in one table)
BLIND SPOTS: <stale snapshot / missing coverage lines>
NEXT: /market screen <mode> to rank a slice; /invest <ticker> to actRules: same citation/no-advice/blind-spot discipline as everywhere else; composite overlay keeps point-in-time honesty (its as-of vs live bars are never merged into one ranking).
MODE REGIME (= current market regime assessment)
Assess where the market stands using the vault's own regime machinery.
READ-ONLY, stdout only. Sources:
wiki/investing/benchmarks/macro-regime-tables.md (Fed plumbing + curve +
credit + yen + VIX percentile; cite its as-of),
wiki/investing/benchmarks/correlation-clusters.md,
wiki/maintenance/calibration/regime-stratification.md (our OWN strategy
hit rates BY regime; cite generation date), plus factors.db MACRO factors
for fresh readings (rate_shock window math fresh from store, no hardcoded
values).
Assessment structure:
== /market regime -- <YYYY-MM-DD> ==
POLICY PLUMBING: WALCL/WRESBAL/SOFR/RRP state vs thresholds
[macro-regime-tables.md, as-of <date>; fresh db readings flagged]
RATES PATH: curve + credit state; rate_shock gate ACTIVE/CLEAR
(computed fresh: DFII10 60-session delta, DGS10 nominal confirm)
VOL STATE: VIXCLS level + percentile; halt/throttle band mapping
CARRY/YEN: DEXJPUS signature present/absent [engine Section 8]
LIQUIDITY VERDICT: quote engine Section 7 VERDICT line verbatim
DEPLOYMENT POSTURE: resulting multiplier stack = MIN(active states)
[doctrine verbatim] -- measured, not recommended
OUR HIT RATES BY REGIME: trend/vol stratification of our own judgments
[regime-stratification.md]
BLIND SPOTS: <stale tables / missing series>
NEXT: /market watch macro for alert framing; /gate for doctrine verdictsRules: quote doctrine thresholds verbatim with attribution; multipliers stack by MIN, never multiplied; no advice leak ("posture is X" is a measurement, never "so you should...").
Cross-mode integration
/brief: daily narrative briefing with signal dashboards; WATCH is the lighter pre/post-close sweep./risk+/portfolio risk: exposure/VaR measurement; SCREEN borrows the HELD flag only./advise+/portfolio advise: allocation/sizing owner; hand it screen tables as the candidate/universe layer; oversold/breakdown outputs are regime inputs./invest <ticker>: the deeper dive; screens seed candidates, never replace it./backtest: validate a screen RULE historically first -- a screen is a today-scan, not evidence of edge./networth: owns holdings/account truth.- Weekly calibration loop: all modes consume the same factors.db the job ingests; rerun after ingest for fresh as-of.
Verification
- Strict YAML parse;
python skeleton/eval/eval-skill-overlay.py --checkexits 0; non-ASCII scan zero hits. - Dry-run each mode end-to-end on the live db; every number traces to a cited source; scored+skipped == 107 on every screen run.
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
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.

