done

name	done
description	Complete a board task — commit, push, update board status, log to CHANGELOG
user-invocable	true
disable-model-invocation	true
allowed-tools	Bash, Read, Edit, Write, Glob

Wrap up a completed board task with full bookkeeping. The argument $ARGUMENTS is the task title or keyword to match on the board (optional — if omitted, infer from recent work context).

Steps

1. Identify the task. If $ARGUMENTS is provided, search the board for a matching item. If not, infer from the current working directory and recent commits what was just completed.

2. Verify the work is done in the sub-repo:

   • git status --short — all changes committed?
   • If uncommitted changes exist, ask the human whether to commit them first
   • If tests exist for this repo, run them and confirm green
   • git push if not already pushed

3. Update the board item:

   "/c/Program Files/GitHub CLI/gh.exe" project item-edit --project-id PVT_kwHOAH3il84BRykI --id <ITEM_ID> --field-id PVTSSF_lAHOAH3il84BRykIzg_g5JA --single-select-option-id 98236657
   

   (Status field → Done)

4. Log to CHANGELOG.md in C:\Users\pon00\Projects\forensic-accounting-toolkit\CHANGELOG.md:

   • Add an entry under the current date section
   • Include: what was done, which repo, specific files changed, test results
   • Match the existing CHANGELOG format

5. Check if ECOSYSTEM.md needs updating in C:\Users\pon00\Projects\forensic-accounting-toolkit\ECOSYSTEM.md:

   • Did a blocker get resolved? (XB-001, XB-002)
   • Did publication status change?
   • Did a backlog item get completed?
   • If yes, update it. If no, skip.

5b. Sync hub CLAUDE.md counts. After updating ECOSYSTEM.md, verify test counts and extractor counts are current:

• Run uv run pytest tests/ --co -q in the completed repo. Parse the test count from the last line.
• Compare to the count shown in hub CLAUDE.md ecosystem table. If different, update the number.
• Also check ECOSYSTEM.md publication table for the same repo's test count — update it to match if different.
• If the repo is krff-shell (delivery shell), also count .py files matching kr-dart-pipeline/kr_dart_pipeline/extract_*.py. If the count differs from CLAUDE.md's "N extractors" claim, update it.
• Only touch CLAUDE.md and ECOSYSTEM.md if a number actually changed.
• Stale name check: After any file edit in this session, run:

  grep -rl "kr-forensic-finance" /c/Users/pon00/Projects/*/CLAUDE.md /c/Users/pon00/Projects/*/README.md /c/Users/pon00/Projects/forensic-accounting-toolkit/*.md /c/Users/pon00/Projects/forensic-accounting-toolkit/*.conf 2>/dev/null | grep -v ".git" | grep -v "reports/"
  

  If any files are returned, replace kr-forensic-finance with krff-shell in those files (excluding historical reports/ directories).

5c. Sync cross-issue files. If the completed task resolves a cross-issue:

• Update cross-issues/XB-NNN.md status from ACTIVE to RESOLVED with date and commit hash
• Example: **Status**: RESOLVED (2026-03-15, commit \abc1234`)`

6. Commit and push the toolkit hub (CHANGELOG.md, ECOSYSTEM.md, CLAUDE.md, cross-issues changes)

7. Report summary:

   • What was closed
   • Board status
   • One-line "Next unblocked AI task: ..." suggestion
   • "Run /triage to rescan for newly unblocked tasks."

8. Content assessment. Review what was just completed and assess whether it involved any of these content-worthy patterns:

   • A plan or approach failed and we pivoted to an alternative
   • A new API endpoint, data source, or technique was discovered
   • A significant metric improvement was achieved (quantify: before → after)
   • A multi-step debugging or investigation sequence resolved
   • A cross-repo integration issue was identified and fixed
   • A domain-specific insight emerged (regulatory, accounting, market structure)

   If NONE match: skip silently. Do not mention content at all.

   If ANY match: suggest ONE specific /capture command at the end of the report. Format:

   Content: This task involved [pattern]. Consider:
     /capture {suggested-title}
   

   Do NOT auto-run /capture. The human decides.

9. Cascade scan. After all bookkeeping, actively search for new work created or unblocked by the completion. Check these 6 sources:

   Source	What to check	How
   Downstream data	Did this change data files consumed by other repos?	Check if any parquets in krff-shell/01_Data/processed/ are newer than copies in kr-derivatives/data/input/. If so: [DATA REFRESH]
   Test ripple	Did downstream tests break?	Run uv run pytest tests/ -x -q in repos that consume outputs from the changed repo. If failures: [TEST FAILURE]
   Convention drift	Did the change introduce convention violations?	Quick check: new files without constants.py patterns, new extractors not counted, missing conftest.py. If drift: [CONVENTION]
   Unblocked items	Did this unblock board items or backlog items?	Check board Todo and ECOSYSTEM.md backlog for items that reference the completed task or its repo. If unblocked: [UNBLOCKED]
   New TODOs	Did the committed code introduce new TODOs?	git diff HEAD~1 HEAD in the repo, grep for TODO/FIXME/HACK. If new: [NEW TODO]
   Cross-issue cascade	Did resolving a cross-issue unblock other work?	Check cross-issues/ for items that reference the resolved issue. If unblocked: [CROSS-ISSUE]
   Known Gaps resolved	Did completed work resolve a Known Gap?	Compare files touched and tests added against the target repo's ## Known Gaps table in CLAUDE.md. If a gap appears resolved: [KNOWN GAP RESOLVED] — suggest removing or updating the row

   Output format (appended to the report):

   Cascade scan:
     [UNBLOCKED] kr-derivatives Run 4 — XB-001 resolved, Run 3 complete
     [DATA REFRESH] kr-derivatives inputs may need sync → bash ecosystem.sh copy-parquets
     [CLEAN] No new TODOs, no test failures, no convention drift
   
   Discovered tasks: 1
     → kr-derivatives Run 4 (resolve 32 remaining >10x moneyness outliers)
   
   Recommended: /work kr-derivatives
   

   If cascade finds nothing: Cascade: clean — no new tasks discovered.

   Skip sources that clearly don't apply (e.g., don't check downstream data if the task was documentation-only). Only run downstream tests if the change modified code or data files, not docs.

10. Lessons check. If the task involved fixing a mistake that the system made (stale data echoed, wrong recommendation, convention violation missed), append a one-line lesson to C:\Users\pon00\Projects\forensic-accounting-toolkit\lessons.md with the date and what was learned. Only add lessons for systemic issues — not one-off bugs. Format:

    - **Lesson title.** Description of the rule. (Learned: YYYY-MM-DD, context)
    

Board Reference

• Project ID: PVT_kwHOAH3il84BRykI
• Status field: PVTSSF_lAHOAH3il84BRykIzg_g5JA
• Done option: 98236657
• In Progress option: 47fc9ee4
• Owner field: PVTSSF_lAHOAH3il84BRykIzg_g-mo
• Fetch items: "/c/Program Files/GitHub CLI/gh.exe" project item-list 1 --owner pon00050 --format json

Rules

• Always ask before committing uncommitted changes — don't assume they're ready
• If tests fail, stop and report. Do not mark the task as Done.
• CHANGELOG entries must be specific: "Fixed X in Y" not "Updated code"
• Do not over-update ECOSYSTEM.md — only touch it if status actually changed