triage

name	triage
description	Aggressively scan all task sources and produce a ranked "do this next" action list
user-invocable	true
disable-model-invocation	true
allowed-tools	Bash, Read, Grep, Glob

Scan the entire ecosystem for actionable work and produce a prioritized task list. Runs triage-scan.sh to collect signals, then applies LLM judgment to rank heterogeneous findings.

Arguments: $ARGUMENTS

• No argument → full scan (all 13 sources)
• quick → fast scan (sources 1-3 only: board, git hygiene, backlog)

Steps

1. Run the scanner:

   bash "C:/Users/pon00/Projects/forensic-accounting-toolkit/triage-scan.sh" $ARGUMENTS
   

   If $ARGUMENTS is empty, run without arguments (full mode).

2. Check for previous triage snapshot (for delta reporting): Read C:\Users\pon00\Projects\forensic-accounting-toolkit\triage-last.json if it exists. If it exists, note the timestamp and previous top findings for comparison.

3. Parse each section of the scanner output and classify findings.

4. Validate board freshness. The board is a claim about what needs doing. The filesystem is evidence of what's been done. Claims go stale. Evidence doesn't. For each board Todo/In Progress item:

   • Cross-reference the item's title against recent commits (shown as ~ lines in GIT HYGIENE). If commits show the described work was already completed, the board item is stale.
   • Check for progress artifacts in the relevant repo — e.g., reports/ directories with run logs, prep files, or lesson files that indicate iterations beyond what the board title describes.
   • Check whether the backlog (ECOSYSTEM.md) describes a later iteration of the same task. If so, the backlog is more current than the board.
   • Mark stale items as [STALE BOARD] in your output. Do NOT recommend them as next tasks.
   • If a board item is stale, recommend updating it (rename or close) as a housekeeping item.

5. Gap synthesis. Go beyond reporting status — discover new work. Using the IMPLEMENTATION COMPLETENESS and STRATEGY ALIGNMENT scanner sections:

   a. Separate blocked from actionable. Stubs tagged [BLOCKED] go to "Blocked (informational)" — never recommend these. [OPEN] stubs are candidates.

   b. Cross-reference against board and backlog. If a gap is already tracked, don't surface it again. Only surface gaps that are NOT already tracked anywhere.

   c. Assess strategic importance. Not every gap is worth filling now:

   • Does this gap block other work? (high priority)
   • Is this gap in a repo that's otherwise "complete"? (high value — finishing repos)
   • Does strategy docs mention this as needed? (alignment signal)
   • Is this a stub in active code vs. dead/unused code? (active = higher)

   d. Generate task descriptions. For the top gaps (max 3), write a one-line task suitable for the board: repo, what's missing, why it matters.

   e. Cross-reference Source 13 (Known Gaps). For each Unblocked Known Gap:

   • Skip if already tracked on the board or backlog
   • Prefer the Known Gap description over raw grep stubs when both exist for the same issue
   • Boost priority for gaps whose status contains "bug" or "risk" suffixes

   f. Output in "Gaps discovered" section. If none: "No new gaps found."

6. Rank findings using this priority matrix:

   Category	Priority	Rationale
   Test failure detected	P0	Broken code blocks everything
   Uncommitted changes	P0	Risk of lost work
   Unpushed commits	P1	Work done but not backed up
   Board P0 tasks (AI-owned, unblocked)	P1	Explicitly prioritized
   Data sync staleness (STALE parquets)	P1	Downstream results are wrong
   Cross-issue newly resolved	P2	Unblocks downstream work
   Board P1 tasks (AI-owned, unblocked)	P2	High priority
   Backlog P1 not on board	P2	Needs board promotion
   Open stub in active code	P2	Completeness of working system
   Strategy-mentioned but untracked	P3	Future alignment
   Known Gap — Unblocked (bug/risk)	P2	Documented gap with severity signal
   Known Gap — Unblocked (other)	P3	Documented gap, no severity signal
   Convention drift (DRIFT)	P3	Important but not urgent
   Code signals (TODO/FIXME)	P3	Technical debt
   CHANGELOG staleness	P3	Documentation hygiene
   Stale branches (>3)	P3	Repo cleanup

7. Output the ranked action list (cap at 8 items). Use this format:

   === TRIAGE ===
   
   Immediate (do before anything else):
     1. [UNCOMMITTED] kr-derivatives — 2 files modified
     2. [DATA STALE]  kr-derivatives — price_volume.parquet older than source
                      -> Run: bash ecosystem.sh copy-parquets
   
   Next task:
     3. [BOARD P1]    kr-derivatives Run 4 — 32 remaining outliers
                      -> /work kr-derivatives
   
   Backlog surfaced:
     4. [BACKLOG P1]  kr-beneish PyPI publication — not on board
   
   Gaps discovered:
     5. [OPEN STUB]   kr-derivatives — repricing engine has NotImplementedError
                      Why: core functionality, blocked only by SEIBRO data
     6. [INTEGRATION] kr-enforcement-cases — enforcement labels not yet consumed by krff-shell
                      Why: violations.csv exists but label pipeline not wired
   
   Blocked (informational):
     - kr-derivatives — SVI fitting: requires KRX options data (Phase 2)
   
   Clean: 6/8 repos clean | Board: 3 AI todo | Last CHANGELOG: today
   
   Recommended: /work kr-derivatives
   

   Grouping rules:

   • "Immediate" = P0 items (must be resolved before doing anything else)
   • "Next task" = highest-priority P1/P2 board item or backlog item
   • "Backlog surfaced" = items found in ECOSYSTEM.md backlog not on the board
   • "Gaps discovered" = new work found by sources 11-12, not already tracked
   • "Blocked (informational)" = stubs/gaps requiring human action, never recommended
   • Summary line at bottom: repo cleanliness count, board todo count, CHANGELOG status
   • Final line: single recommended next command

8. If previous triage exists, add a delta section:

   Delta (since last triage YYYY-MM-DD HH:MM):
     [RESOLVED] kr-beneish — was UNCOMMITTED, now clean
     [NEW]      kr-derivatives — STALE parquets detected
   

9. Write triage snapshot to C:\Users\pon00\Projects\forensic-accounting-toolkit\triage-last.json:

   {
     "timestamp": "2026-03-15T14:30:00",
     "mode": "full",
     "findings": {
       "uncommitted": ["repo1"],
       "unpushed": ["repo2"],
       "stale_data": ["price_volume.parquet"],
       "board_todo": 3,
       "backlog_open": 5,
       "code_signals": 12,
       "convention_drift": [],
       "cross_issues_active": ["XB-002"],
       "open_stubs": 3,
       "blocked_stubs": 1,
       "strategy_untracked": 9
     },
     "known_gaps_unblocked": 0,
     "top_task": "kr-derivatives Run 4",
     "recommended_command": "/work kr-derivatives"
   }
   

Rules

• Read-only. Do not fix, commit, push, or modify anything. Just report.
• Cap output at 8 ranked items. If more exist, mention count: "... and N more P3 items"
• If scanner fails entirely, report the error and suggest bash ecosystem.sh status as fallback.
• If gh CLI is offline, the board section will say UNAVAILABLE — continue with other sources.
• Be concise. The user wants a 15-second scan, not a 5-minute report.
• Always end with a single Recommended: line — the one command to run next.