Part of #63 (Phase 1).
The problem
The workplan-* family is five verbs and two skills, and three of those verbs plus one whole skill have never run. Validated so far: update and resume, from an installed plugin on 2026-08-20 and again on #25 on 2026-08-25. Never run: create, read, close, and /qe:workplan-project against a real report bundle.
That matters more here than it would for a read-only skill. These verbs file issues, edit bodies in place, and close-and-succeed plan chains — close in particular creates a successor, posts a ledger citing it, and closes the old plan in an order that is meant never to leave the chain dangling. A procedure that has never executed is a hypothesis, and this family's hypotheses write to GitHub.
The work
workplan-project has a candidate waiting: ~/work/quantecon/_audits/2026-08-18-quantecon-py-audit/, named in #25 as the richest available bundle. Run it end to end and record what the skill did against what its SKILL.md says it does.
create, read and close are best exercised as a chain rather than in isolation, since close builds its successor through create's succession source — the carry-forward register, never the old body. This tracker's own lifecycle is one honest venue for that.
Record each run the way the audit family does: a file under reviews/, following the shape of audit-run-meta-2026-08-25.md — what was invoked, what it produced, which claims in SKILL.md held, which never came up, and which the run had to work around. Only the third is a defect. Findings that are defects become their own issues, per the family's own convention.
Acceptance criteria
Part of #63 (Phase 1).
The problem
The
workplan-*family is five verbs and two skills, and three of those verbs plus one whole skill have never run. Validated so far:updateandresume, from an installed plugin on 2026-08-20 and again on #25 on 2026-08-25. Never run:create,read,close, and/qe:workplan-projectagainst a real report bundle.That matters more here than it would for a read-only skill. These verbs file issues, edit bodies in place, and close-and-succeed plan chains —
closein particular creates a successor, posts a ledger citing it, and closes the old plan in an order that is meant never to leave the chain dangling. A procedure that has never executed is a hypothesis, and this family's hypotheses write to GitHub.The work
workplan-projecthas a candidate waiting:~/work/quantecon/_audits/2026-08-18-quantecon-py-audit/, named in #25 as the richest available bundle. Run it end to end and record what the skill did against what itsSKILL.mdsays it does.create,readandcloseare best exercised as a chain rather than in isolation, sinceclosebuilds its successor throughcreate's succession source — the carry-forward register, never the old body. This tracker's own lifecycle is one honest venue for that.Record each run the way the audit family does: a file under
reviews/, following the shape ofaudit-run-meta-2026-08-25.md— what was invoked, what it produced, which claims inSKILL.mdheld, which never came up, and which the run had to work around. Only the third is a defect. Findings that are defects become their own issues, per the family's own convention.Acceptance criteria
/qe:workplan-projectrun against the QuantEcon.py audit bundle, with a record underreviews/create,readandcloseexercised, with a record underreviews/