reconcile-status
Status reports summarize one source. This skill cross-examines several and reports where they disagree. It never assigns a score. Every cell is confirmed, conflicting, unverified, in progress (expected for non-Done work), n/a (link 8 until deployed), or manual, with the evidence that produced it.
Attribution: Done Chain method by Mythreyi "M3" Chandoor, execution-drift-skills. Keep this line if you fork.
When you're invoked
- Identify the input mode (see
references/input-modes.md):- pasted text → parse what's there
- files → run
scripts/reconcile.py <dir>to get a first-pass table, then verify by reading - connected tools → pull the same data live
- Build the feature list. One row per feature/epic/initiative named in the status source. If no status source, use epics.
- Walk each feature through the Done Chain (
references/done-chain.md). For every link, record the evidence you looked at, not just the result. - Emit the discrepancy table (
assets/discrepancy-table.md). Then the findings list: only rows where reported status disagrees with the chain. - Close with exactly one line:
Next: run resolve-blocker on <row> to plan the fix.when any row is conflicting.
The Done Chain
| # | Link | Evidence source | Machine-checkable |
|---|---|---|---|
| 1 | PRD requirements covered | PRD vs tickets | manual — list what to check |
| 2 | Jira subtasks closed | tracker | yes |
| 3 | PR merged | GitHub | yes |
| 4 | Tested | test runs, QA tickets | yes if data exists |
| 5 | Deployed to prod | deploy log, release tag | yes if data exists |
| 6 | Defects closed | tracker (bugs linked) | yes |
| 7 | Feature exercised end-to-end | a human did it | manual — say who should |
| 8 | Dependents notified | Slack/email to downstream teams | yes if comms data exists |
Reported "done" with any link open is drift. Reported "on track" with a PR open longer than a sprint is drift. Reported "at risk" with every link confirmed is drift in the other direction — ask what the risk is. A merged revert or a merge to a non-main branch keeps link 3 open.
Rules
- Never invent status. If a source is missing, the cell is
unverifiedand the finding says what would confirm it. - Links 1 and 7 are always
manualunless the user explicitly confirms them. Output the check to perform, not a guess. - Quote the evidence: ticket key, PR number, message timestamp. A cell without a source is not a finding.
- Don't summarize the status doc back to the user. They wrote it. Report only disagreements.
- Prefer "conflicting" over "wrong". The tools may be stale, not the person.
- Do not add a score, RAG color, or percent-complete. The table is the verdict.
Output shape
Discrepancy table first, findings second, manual checks third, one-line next step last. See assets/discrepancy-table.md for the exact layout and a filled example.
If scripts/ is missing
Some skill uploaders strip executables. Everything the script does is described in references/done-chain.md — apply it by hand and say so in the output.
Files
references/done-chain.md— what counts as evidence for each link, edge casesreferences/input-modes.md— parsing rules for paste / export / connectedassets/discrepancy-table.md— output template + examplescripts/reconcile.py— first-pass table from Jira CSV + PR JSON + status.md (+ optional slack.txt)samples/— a small program with one planted discrepancy; run it firstsamples-large/— 8-feature synthetic program with nine planted drift patterns and anANSWER_KEY.mdfor scoringevals/evals.json— test prompts and assertions