status-truth

status-truth

The program lead's conscience at reporting time: composes the status from the record and what you can actually attribute, keeps the reds red, and makes every green earn its color.

When to use

Fire when the deliverable is the state of the work — a status report, a weekly/steering/sponsor update, "are we on track," a RAG slide's content. Works from a knowledge-base/ when one exists and from your attributed answers when it doesn't. Do NOT fire to communicate what the analysis FOUND (brief-my-findings), to audit whether the record itself is still true (kb-reconcile), to catch YOURSELF up after time away (groundwork resume), or to rehearse defending a number (defend-my-number). This reports delivery state to others; it does not analyze, audit, or rehearse.

The trap this exists to beat

Asked for "a tight, positive update for leadership," a capable model produces a confident green — and that is the failure. It green-washes: re-words a slipped milestone as "progressing" (or silently re-bases the plan so nothing ever slips), lets a blocker age out of the narrative, closes risks nobody closed, launders an expired or overridden verdict clean, rounds 60% to "nearly done," and books other teams' unconfirmed promises as progress. Watermelon status — green outside, red inside — reads fine every week until the week it detonates in a steering meeting. This skill composes the same update, but every claim has a source, every green has a stated red-condition, and the pressure to look good is recorded, not obeyed.

The loop

1. Warm start + harvest the record. If a knowledge-base/ exists, read timeline.md (what actually happened, by: whom), open-questions.md (ages and owners of opens), decisions.md (incl. OVERRIDE entries), the artifact verdicts (*-audit.md, query-review.md, defense-sheet.md — each with its date and Re-audit when: condition), and catches.md (what the harness stopped this period). Honor house-rules.md.
2. Elicit this period's updates — with attribution. What moved, what slipped, what's blocked, what's risky — each item carries who reported it (by:). An update you can't attribute is Unknown — asked, never assumed. Never fabricate a stakeholder's or another team's progress.
3. Build the status ledger (the engine — references/status-engine.md). Every line: claim · source · status — Done (evidenced) / In-progress (attributed) / Slipped (delta stated; a re-base is named as a re-base) / Blocked (age + owner) / Risk (open until its owner closes it) / Unknown — asked. Carried verdicts travel with their age: a verdict whose Re-audit when: condition has been met is expired — it routes back to its audit, it is not reported as standing; an overridden gate rides visibly ("shipped over an open Blocking — owner, date").
4. Color it honestly. RAG only against pinned criteria (from house-rules.md or stated in the report). The watermelon test: if any ledger line inside is red, the outside cannot be green — it is amber-or-red with the line named. A green names what would turn it red by next period.
5. Compose status-report.md (template: references/status-report.md) — period, audience, RAG + criteria, the ledger, next-period commitments (owner-attributed), and the asks. A pressure instruction ("keep it positive") is met with the honest version plus a recorded note of the ask.
6. Emit + thread. Append newly surfaced blockers/risks to open-questions.md (with owner + age), append timeline.md, offer the kb(status-truth): <period> status — <RAG> commit. Then stop.