[shafty023]

Purpose

One release (vibe 1.0.67) for the vibe lessons ledger entries added on 2026-10-20 after #317 merged, headed by the operator's hard rule on local paths. Successor to #317; later lessons batch into this PR until it merges.

Lessons carried

Hard rule: no local paths in anything written for people (operator, 2026-10-20). Tickets, PR bodies, PR comments and review replies never reference a path on the operator's machine (home, temp, scratchpad, worktree or .control/ paths) or say something is on someone's laptop. Screenshots are uploaded and embedded (create-document-version inlineImages), text evidence is quoted inline, briefly. Extension: a plan the ticket references is attached, never pointed at by path; an implementation plan is the PLN linked with create-artifact-link (PRODUCES), and a QA plan or other working file is uploaded with upload-attachment and named by its attachment name.

• fleet-orchestrator/SKILL.md: new "Artifacts people read" section.
• fleet-orchestrator/references/templates.md: the worker preamble and the Codex resume prompt carry the rule, so every writer gets it.
• vibe/references/ticket-template.md: the live ticket rules (owner for ticket and PLN text); Visual evidence and Running app checks now carry embedded captures instead of .control/ artifact paths, with the ticket worker and handoff SKILL.md updated to match.
• agents/vibe-environment-worker.md: the PR body and review replies follow the same rule.
• Mechanical check: publishable-text.mjs pointerProblems refuses a line naming a local absolute or home path, a .control/ capture or the local plans folder, never echoing the value. live-ticket-check.mjs reports each as a Body problem; pr-facts.mjs adds each to blocking.prBody. The new tests fail on a home-directory screenshot path, a tilde path, a .control/ capture and a local plan path.

Codex to Claude takeover (#8958 preview rebuild refused). The takeover preamble said vibe could not move a Codex session, so the Claude writer ran beside a ledger that still named the dead Codex writer and dispatch-inputs refused every environment request. vibe 1.0.66 already has the harnessSwitch migration; the orchestrator skill and takeover preamble now use it before any other step. A seeded or blank mode change for a rebuild is touch --mode plus a new environment request (environment.md).

Orchestrator rules (fleet-orchestrator SKILL.md and judgment.md).

• Preview fixes: a fix merged to main reaches a preview only once the preview's deploy ref contains it (stage deploys from green); check ancestry first.
• Stale pre-enqueue worker (#8959): stop the worker and its processes rather than messaging it, relaunch narrowly; before enqueue run only the lanes the diff can affect, not full suites.
• Storybook version bump conflicts (#8900): the shared release bump is made only at enqueue, with auto-merge armed in the same minute.
• ISS-12644: being on main is not evidence of the final design; ask the owner which side is scaffold.
• #8900 announcer: when a worker's alternative is accepted, withdraw the instruction at once.

Not plugin changes (left out): the remote machine running vibe 1.0.54 (an upgrade, not a rule), and the #8953 drag-to-status note already encoded in #317.

Cost

• Instruction text (vibe agents/*.md and skills/**/*.md): +46 / -20 lines, net +642 words, mostly the hard rule in its four owners.
• Whole PR: 18 files, +108 / -26.

Commands run (final head)

• uv run pytest plugins/vibe: 361 passed
• node --test over all 48 vibe *.test.mjs files: 438 passed
• uv run ruff check and uv run pyright on the changed Python test: clean

Version

vibe 1.0.66 to 1.0.67 (both plugin manifests and the README), one CHANGELOG entry.