Skip to content

v2.4.0 — sync to parley-deck-cli 1.39.0

Choose a tag to compare

@feci feci released this 06 Aug 13:04
· 12 commits to main since this release

Sync to parley-deck-cli 1.39.0. That release changed what decides whether a headless participant can write its own artifact: headless_args is the launch, and autonomous_write became a declaration audited against it. The skill still taught the older two-list model — guidance that would reproduce the defect 1.39.0 fixed.

Fixed

  • SKILL.md named the wrong source of truth. "The source of truth for each agent's mode is the spec's autonomous_write field" pointed readers at the field that launches nothing, three lines under the table a facilitator reads to pick a flag. The source of truth is now the effective launch argv; autonomous_write is a verification contract to audit it against. A config layer that overrides headless_args replaces it wholesale and can silently drop the enabling flag.
  • The command-construction recipe described an assembly the CLI does not perform. Split into an explicit manual facilitation branch and a Parley CLI branch, where resolved headless_args is the complete argv template and nothing is appended afterwards.
  • writeModeArgs removed from the documented config shape, with a migration note: merge its arguments into headlessArgs and drop the field.
  • compatibility.json skillVersion was 1.4.3 while the package was 2.3.0 — four releases of silent drift, because nothing read the field.

Added

  • opencode row in the Autonomous Execution table.
  • A version-sync guard in both scripts/build-addon-manifest.js and the test suite. It lives in both on purpose: prepack runs the script but not node --test, so a test-only assertion would leave npm publish ungated — the exact path the drift took.

Designed and reviewed through a full Parley Deck run (skill-sync-cli-1-39): four participants, two design rounds, two consensus revisions, one review round plus a fix-up cycle. One MAJOR was caught in review. Three of the four defects were found by participants rather than by the brief.