One brief per initiative, named EXECUTE-<topic>.md.
Keep its path stable while the work remains open.
Each brief owns its remaining work, acceptance checks, and dated evidence.
| Brief | Remaining question |
|---|---|
| GitHub pivot | Can the sender deliver a digest beyond Verified Destination Addresses? |
| Homepage rebuild | Does a fresh 30-Skill sample pass classification review? |
| SEO keyword rework | Do the production research pins meet the admission bar? |
| Cost follow-up | Do new measurements meet the recorded cost and D1 targets? |
| SEO recovery | What does the weekly panel show before the November gate? |
| Public API v1 | What remains outside the published contract? |
| Skillgen bootstrap | How many model calls does a GLM 5.3 bootstrap run need? |
VISION owns product scope. Architecture and ADRs own system decisions. SEO recovery owns indexing experiments, measurement dates, and the gate. Its panel is also read by the daily check-in.
Comparison research and the comparison brief retain sources behind published comparisons. The editorial evidence retains sources, replay observations, and claim limits behind published articles.
Give each brief a title, next action, Done means: acceptance check, ledger, and dated evidence log.
Record observations with their date and source. Query GitHub or production for current delivery state.
Do not copy product strategy, schemas, or runbooks into a brief.
After acceptance, remove a completed brief unless it owns evidence that another document or published article needs. Git history retains its implementation record.