[JacobStephens2]

Summary

thirdshift skill <skill> [issue <issue>]
  resolve exact Factory/repository name, then Factory shorthand
  create temporary worktree at the checkout's committed HEAD
  link Factory skills; hand over the optional issue
  run one session on the default Harness, Model and Effort
  require terminal completion and a nonblank final message
  print final message; retain local work or remove the worktree

Adds CLI help and manual documentation. Repository skills resolve by directory
name under .agents/skills/ or .claude/skills/; unknown names list available
skills without starting a session. Uncommitted checkout changes stay outside the
session. Standalone skills create no Command record or Session log.

Closes #965.

Evidence

• Before: the retained a_codex_stream_ending_before_turn_completion_is_not_success
  regression failed: a zero-exit Codex stream without turn.completed printed
  “I will run validation now.” and returned 0 despite unfinished work.
  After: the exact regression passes; partial Muse and OpenCode messages also
  fail with an early-ending cause and empty stdout.
• The focused checks passed: 168 Harness unit tests, 47 CLI/Harness integration
  tests, and 11 release-summary tests. These cover exact-name/shorthand lookup,
  one-session execution, configured Harness settings, issue handoff, checkout
  isolation, worktree retention/removal, failure causes, and CLI help.
• CARGO_TARGET_DIR=/home/jacob/repos/thirdshift/target /usr/bin/time -p scripts/affected-tests main
  passed on 29963ff6: 2,366 Rust tests (57 skipped) and 36 browser tests.
  Total wall time: 1,150.34s, including browser checks and 21.76s of incremental
  Rust compilation. Nextest was unavailable, so the selector used cargo test.
• cargo clippy --all-targets -- -D warnings and cargo fmt --check passed.

Test seams

• The thirdshift skill CLI with real Git repositories and fake Harness
  executables: skill lookup, prompts/final output, checkout isolation, worktree
  retention, session completion, and failures.
• Existing CLI help, Harness selection/interpretation, worktree ownership and
  release-summary interfaces provide regression coverage.

Review

• Standards: one possible Duplicated Code finding, fixed by sharing the
  read-only worktree cleanliness predicate while retaining caller-specific
  ownership checks and Git execution.
• Spec: two P2 early-ending findings (Codex without a terminal event, and a
  resumed Claude root turn without its next result). Both exact
  regressions failed as reported, remain in the suite, and pass after the fixes.
• Both reviewers read all 32 changed files, including generated
  site/docs/index.html. No changed file was left unread and no generated output
  was excluded. Standards covered the amendments through 9af4d969; the Spec
  reviewer also inspected the final Claude resume correction in 29963ff6.
  Reports and summary.json are retained with this Session trace.

Unaddressed findings

Standards

None.

Spec

None.

Security

None.

Security review outcome

No unaddressed introduced findings.

Merge Danger

Door: two-way

Blast Radius: CLI

Adds standalone skill execution through the existing Harness adapters. Terminal
completion checks also apply to release summaries; regular Run completion policy
is unchanged. Worktree cleanup shares the existing cleanliness and ownership
checks.

Built with codex codex-cli 0.164.0 · gpt-6.1-sol · xhigh