Repository navigation
Description
Parent
Refs #15659 (the map decision this implements). First of five alignment items that bring app-control closer to Lauren Tan's verification skills (cursor/plugins pstack/skills/create-verification-skill and maintain-verification-skill, MIT).
Depends on
- none: on the frontier now
Outcome
A control skill's feature map says what proves each feature works. Every drive step states the action, the exact command or handle, and the result a user should observe, and each feature's steps end with a proof step naming the end state and the captures that show it. A live check can then judge a feature from the map alone, with no change to read an expectation from.
Demo: open any feature file in this repository's control skill; each step shows its expected result and the last step is the proof. Run /app-control:map check; a step with no stated result fails it. A drive run's report shows each step's stated result beside what was observed.
Her wording, kept: each feature file answers "what the feature is, how to reach it, how to drive it with the harness, and what observable end state proves it works" (create-verification-skill/SKILL.md:36); the drive section "uses labeled bullets that pair each user action with an exact command and observable result" (references/feature-map-example/README.md:39), ending with a **Proof.** bullet.
Decisions
- Owner, 2026-11-12: "we need to align closer with what her skill actually does, what it produces, how the features are mapped, and how the app is driven and maintained."
- The result stays with its step; no separate proof section. A step and what it should produce change together.
/app-control:drivestill never judges. It records each step's stated result beside what it observed without comparing them, the way it already records a budget value beside its threshold./testing:run-e2estarts each entry point's expected outcome from the feature's stated results and proof step, and adds the change's requirement on top.- This is a breaking change to the control-skill contract (a minor version while the plugin is below 1.0). Existing control skills are refreshed in the same change: this repository's two (about 104 feature files, most of which hand off to two driving-tool features, so the edit concentrates there).
Key interfaces
- The contract's feature file section for drive steps, and the meaning of a
resultin a project's steps file. - Drive's evidence report gains the stated result beside the observed one.
- The contract check gains one rule.
Done when
- The contract requires action, command or handle, and observable result on every drive step, and a final proof step naming the end state and its captures.
-
/app-control:mapwrites them on create, andrefreshtreats a missing or wrong result as doc drift. - The contract check fails a step with no stated result; a test covers it.
- Drive's report carries each step's stated result beside the observed one, with no word that rates the outcome.
-
/testing:run-e2ereads the stated results as the starting expectation. - This repository's control skills are refreshed and pass the check (dogfood), named in the pull request body.
- The pstack upstream record's row for the verification pair is updated.
- Changed skill text ships on a two-version compare of the changed skills' cases (
opusandsonnetaliases, three runs per side,--casefilter, a--max-cost-usdceiling, no new batch after a usage-limit error), run in the one local eval-runner session; the pull request carriesdo-not-mergeuntil the record lands.
Out of scope
- The other four alignment items (preconditions and proof per kind of action; launch and readiness; upkeep; hub conventions and the driver CLI).
- Her output path, her fixed hub section set, asking which skill to refresh when unattended, and a reset step outside drive: each was rejected on record and stays rejected.
Metadata
Metadata
Assignees
Labels
Type
Fields
Work status
Priority
Activity
melodic-automation-lanes commentedon Nov 13, 2026
No work-resolve-issue run started in the 1 hour since
agent-readywas last added. work-unblock removed and addedagent-readyback to start a fresh run on main's workflow files. If this run ends the same way, it re-fires the issue again, waiting longer each time.melodic-automation-lanes commentedon Nov 13, 2026
No work-resolve-issue run started in the 1 hour since
agent-readywas last added. work-unblock removed and addedagent-readyback to start a fresh run on main's workflow files. If this run ends the same way, it re-fires the issue again, waiting longer each time.melodic-automation-lanes commentedon Nov 13, 2026
No work-resolve-issue run started in the 1 hour since
agent-readywas last added. work-unblock removed and addedagent-readyback to start a fresh run on main's workflow files. If this run ends the same way, it re-fires the issue again, waiting longer each time.github-actions commentedon Nov 13, 2026
Size: Large
Complexity: Medium
Uncertainty: Known
Flags: none
Blocked by: none
I read this as a change to the app-control contract. Every drive step would state its action, command or handle, and observable result. Each feature would end with a proof step. The map create and refresh flows, the contract check (with a test), the drive report, and /testing:run-e2e would all follow that contract. This repository's control skills (about 104 feature files) would be refreshed, and the pstack upstream record updated. The done-when items and the owner decisions are clear, so I found nothing missing. The work spans several subsystems and needs a local eval run, so it is large and probably needs splitting. I found no duplicate among the open issues and no merged pull request that already covers it.
Automated triage. A maintainer reviews it.
github-actions commentedon Nov 13, 2026
The triage sweep dispatched intake triage for this issue (dispatch 1; 2 runs that reach intake, or 6 dispatches, hand it to a maintainer).
melodic-automation-lanes commentedon Nov 13, 2026
Not admitted to the issue lane yet: sub-issue #15913 is open.
melodic-automation-lanes commentedon Nov 13, 2026
Split at intake (owner Q93). This issue is Size Large, so its done-when is split into 6 sub-issues, each holding its task-list lines as written.
Automated split by the lanes App. Admission waits on this issue while a sub-issue is open; work-unblock closes it as completed once every sub-issue is closed as completed by a merge or a listed human and its done-when is covered.
melodic-automation-lanes commentedon Nov 15, 2026
Every sub-issue of this split is closed as completed by a listed human or a merge (#15908, #15909, #15910, #15911, #15912, #15913), so the lanes App closes it as completed.
chore(app-control): merge main into feat/15798-control-skill-launch-m…
docs(upstream): re-point stale line citations in the pstack row