G
YouTube TikTok Games
Skip to content

feat(pstack): extend statechart with workflow verification - #1233

Open
jgustave wants to merge 1 commit into
cursor:mainfrom
jgustave:feat/workflow-verification
Open

jgustave wants to merge 1 commit into
cursor:mainfrom
jgustave:feat/workflow-verification

Conversation

@jgustave

Copy link
Copy Markdown

Statechart makes broken state transitions visible. This adds our workflow-verification skill as an extension of that concept: an agent-guided workflow linter that traces the contracts through the real implementation and chooses executable checks for the failure paths.

We have found this a more useful extension of the statechart idea for day-to-day workflow review: a model can be internally consistent while the producer, consumer, persisted state, or recovery path disagree. This skill checks those boundaries and brings in statechart when ordering, concurrency, or loops benefit from a bounded model. It requires predicted bug paths to be replayed against the code and distinguishes confirmed findings from untested hypotheses.

The skill covers phase contracts and completion semantics, one maintained definition per contract, producer/consumer compatibility, full lifecycle and continuation tests, interruptions, payload/history pressure, and provider boundary checks. Review requests stay read-only; implementation and live-system permissions remain with the owner. This is an agent skill that guides verification, not a new deterministic CLI linter or a claim that prose enforces correctness.

Adds discoverability from statechart, the skill catalog, and poteto-help, plus the required generated Claude Code tree. No statechart runtime changes or new runtime dependencies. The contributed skill retains its MIT license.

Validation:

  • Plugin definitions, generated harness freshness, pstack documentation checks, skill frontmatter validation, and git diff --check: passed.
  • Existing CI script suite on Node 22 with TMPDIR=/private/tmp: 267 passed, 0 failed, 67 skipped (TLA+ JAR unavailable). The first run with the macOS default temporary path hit the unchanged symlink-launcher test; using the canonical path resolved it.
  • Independent adversarial review: no actionable findings; source and generated copies agree.
  • No installed Cursor/Claude behavioral evaluation was performed; these checks validate packaging and the existing scripts, not agent effectiveness.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant