Repository navigation
docs(rfc): upkeep guidance for a crewmate's Dashboard fields - #19043
Conversation
First Principles Review (Opus 5.5) — 🟡 CONCERNSPremise-level review of First-Principles-Verdict: CONCERNS The RFC lands as Not justified as shipped
What this change shipsInventory (6 items) — 4 justifiedIntent: record the design decision for telling crewmates how to keep their Dashboard fields current. ADDITION (docs only).
[FIRST-PRINCIPLES-REVIEWED] 3877be7 Description read[DESCRIPTION-READ] cf3a774080a962ccff9ef6829ae5228841c06525d8818b98b33c9a6284597f32 That sha256 names the PR description this verdict read. Recompute it with |
UX Review (Opus 5.5) — ✅ PASSUX-level review of UX-Verdict: PASS Nothing to check. [UX-REVIEWED] 3877be7 Description read[DESCRIPTION-READ] 3d030657cc8bbec968a9954841880951ac04351b6295b69149a2ccc53e3b00c2 That sha256 names everything this verdict read: the PR description, plus the 7 evidence file(s) the workflow collected from it and handed this review. It covers that evidence because the media strip replaces every attachment URL with the same placeholder, so an edit swapping one attachment for another would leave the description's captured text identical while what this review judged changed. Recompute it with |
4c7db82 to
6710ece
Compare
GPT 6.1 Review — ✅ no blocking findingsGPT 6.1 completed its review of This comment is updated in place on each push. Review detailsNo findings. False positive or not applicable? A repository writer can comment: |
Design Review (Opus 5.5) — 🟡 CONCERNSDesign-level review of Design-Verdict: CONCERNS The helper copy in §4.3 is the only reason §4.4's sync machinery exists, and it puts filesystem writes into the Watch
[DESIGN-REVIEWED] 3877be7 Description read[DESCRIPTION-READ] d7ca7c99e7d2a1bf5a845caf2fde9e56dd26d8a040411d244aba8d8d3918cd18 That sha256 names everything this verdict read: the PR description, plus the 1 evidence file(s) the workflow collected from it and handed this review. It covers that evidence because the media strip replaces every attachment URL with the same placeholder, so an edit swapping one attachment for another would leave the description's captured text identical while what this review judged changed. Recompute it with |
Opus 5.5 Review — ✅ no blocking findingsReviewed Verdict parsed from the review's SHA-scoped output markers for commit False positive or not applicable? A repository writer can comment: |
6710ece to
dc4c8db
Compare
dc4c8db to
ae8648c
Compare
ae8648c to
1cb412e
Compare
d17a774 to
43f5849
Compare
|
Self-review (crew-pr-reviewer, read-only): CHANGES NEEDED This RFC records the decision to give each crewmate a generated upkeep skill for its agentic Dashboard fields. The motivating gap is real: the |
|
Converted to draft by crew-pr-reviewer: the self-review above is CHANGES NEEDED. Mark ready again (gh pr ready 19043) once the points are addressed. |
43f5849 to
7200611
Compare
|
|
|
Reply to the crew-pr-reviewer self-review (CHANGES NEEDED), addressed in
The RFC index sentence records the 2026-10-12 revision. |
|
Pure RFC changes, which gives the crew mate instructions to update fields in the dashboard/panel. |
Problem / Motivation
Goal: Record the decision that every crewmate with a Dashboard page is told, by
dashboard_fields, what each field only it can fill means, when to write it and how, with a generated helper script where the data lives outside the crew log, named by one pointer line that survives compaction and readable on the profile.A crewmate that adopts a Dashboard page is told which fields are its own and never what they mean, when to write them, or how to get the data. The whole instruction is the
[DASHBOARD]block fromdashboard_agentic.turn_block: the template id and "Fields you write: ...". So fields go unwritten, or stay stale until the person chases them. A field likeciis worse: its data is on GitHub, and the crewmate works out theghcalls and the payload shape every time.[CURRENT ASSIGNMENT](members/<slug>/briefing.md) cannot carry this: it is agent-written, may be empty, and a crewmate with no file tools cannot write it. The person also cannot see what the crewmate is told about its page.Why it matters
The Dashboard tab is what a person reads when they are not reading the thread. A stale field there is a wrong answer shown with authority, and today the only fix is the person noticing and asking.
The implementation adds a line to every crewmate's prompt and a section to the profile. The First Principles lane reads that decision off the base branch, so it lands here first.
Not a goal
docs/request-for-change/and nothing else.{fold, path}).panel_publish). fix(crew): draw the conductor's drawer board from its work ledger #18957 makes the conductor's drawer fold-backed.What changed (motivation → approach → change)
New
docs/request-for-change/rfc-crewmate-dashboard-upkeep-skill.md(statusin-progress) and its row in the RFC index. It records the product owner's ruling of 2026-10-11, revised in review on 2026-10-12, and replaces this PR's first draft (an[OPERATING GUIDE]text block).{"fold", "path"}field is filled by the gateway from the crew log, in both formats:dashboard_templates/manifest.py_fieldandartifact_store/dashboard_package.py_validate_source. So the gap is exactly the{"agentic": true}fields.upkeep: {means, when, how, helper?}. Every built-in agentic field gets one.project-report.ciandpr-watch.lanesname a helper.dashboard_fields(§4.2). Each agentic row of thedashboard_fieldsreply (dashboard_agentic.fields_for_agent) gains anupkeepobject, read from the shipped catalog when the tool is called, never from the crewmate's writable copy. It is a tool reply, so a crewmate with no file tools reads it too, and nothing is generated for the text.builtin/<id>/helpers/<file>under.dashboard-upkeep/<slug>/scripts/, the only thing in that directory. It runsghand prints a readydashboard_writepayload. Only shipped built-ins may name one. A crewmate with no shell writes the field from what it can read instead. A test checks each helper's output passescheck_write.dashboard_upkeep.sync(slug)copies the helpers, writing only what differs, frominstance._commit(adopt, apply, rollback, edit) and from the[DASHBOARD]build (default page, gateway upgrade). Those are every trigger that changes which helpers apply; a v3 package names none.[DASHBOARD]naming the upkeep text indashboard_fields. It rides the member section's three legs (fresh, post-compaction, slim resume). Apply and rollback replies carry it too.GET /api/members/{slug}/briefinggainsupkeep: {pointer, fields, scripts}; the Notes tab shows it read-only.SKILL.mdper board (a crewmate with no file tools cannot open it); one generic manifest-driven update script; a script per board; writing upkeep into[CURRENT ASSIGNMENT].Backwards compatibility
Compatible: documentation only.
Removes nothing: the diff adds one document and one index row.
Tests
N/A — documentation only.
python3 scripts/docs_lint.pypasses.Manual verification
N/A — documentation only. Every symbol cited was read at
a84f616557and re-read at0cda1525d8.Related Issues
Part of #19481 (the implementation PR in the RFC's §7 step 2).
Checklist
Contribution License Agreement
Contributions are accepted under the project's Apache-2.0 license.