docs(autonomy): correct the goal glossary row's terminator claim

[kyle-sexton]

No linked issue.

Doc-alignment fix batch. Version claim: autonomy 0.12.3 pending rebase check -> claimed 0.12.2 (allocated before #1934's 0.12.1 merged; CHANGELOG ordering to be confirmed by the gate).

Ruled SHIPPABLE in-session: live goal.md L59 is a closed two-item terminator enumeration (condition met, /goal clear); zero budget-cap vocabulary on the page; --max-budget-usd is print-mode-only and kills the process while L110 restores the goal on resume. Replacement cell grounded verbatim in L59. Awaiting the orchestrator gate before merge.

Related

• Campaign ledger: Doc-alignment roster-tail campaign — durable save-point (2026-08-04 clean-stop) #1941

🤖 Generated with Claude Code

https://claude.ai/code/session_01WVbP5JXnEaieMgKBRV1PS7

[cursor]

Bugbot is not enabled for your account, so this pull request was not reviewed.

Enable Bugbot in the Cursor dashboard to get automatic reviews on future PRs.

[claude]

Claude finished @kyle-sexton's task in 3m 24s —— View job

---

Code review complete

• Gather context (diff, changed files, CLAUDE.md/REVIEW.md guidelines)
• Understand the request (code review of this PR)
• Review plugins/autonomy/reference/routines.md glossary fix
• Review plugins/autonomy/CHANGELOG.md entry
• Review plugins/autonomy/.claude-plugin/plugin.json version bump
• Post final review

Scope: this PR touches documented harness behavior (the /goal command's termination semantics), so I re-verified the claims against a live fetch of the official docs this session rather than trusting the PR description's account of them.

Verified accurate

• routines.md#L48 fix is correct. I fetched https://code.claude.com/docs/en/goal.md directly: it states "A goal keeps running until the condition is met or you run /goal clear" — a closed two-item set, with zero mention of a budget cap anywhere on the page. The old row's "or a budget cap trips" clause was fabricated; the replacement ("or the goal is cleared") matches the source verbatim.
• CHANGELOG.md 0.12.2 entry claims check out. The --resume/--continue restoration claim matches the doc's "Resume with an active goal" section verbatim ("A goal that was still active when a session ended is restored when you resume..."). I also confirmed --max-budget-usd is documented as print-mode-only via the CLI reference.
• Version sequencing is correct. origin/main's plugin.json is currently at 0.12.1 (the docs(autonomy): document the lane-stop gate's deliberate server-managed settings exclusion #1934 claim referenced in the PR body has since merged), so this PR's 0.12.1 → 0.12.2 bump and the CHANGELOG heading order (0.12.2 > 0.12.1 > 0.12.0) are both consistent post-rebase — the "pending rebase check" the PR body flagged has resolved cleanly, no leftover conflict.

One nuance worth a look (non-blocking)

CHANGELOG.md#L15-L18 states "the cap ends the process while the goal outlives it." The official CLI reference's exact wording for --max-budget-usd only explicitly documents subagent-level enforcement: "spawning another subagent fails with Budget limit reached, and Claude Code stops background subagents that are still running." It does not explicitly state that the top-level/main process exits — that's a reasonable inference from the flag's lead-in ("before stopping"), and given main-turn spend counts toward the same cap it's very likely correct, but I couldn't find an official sentence that says the main process itself terminates. Consider either softening the phrasing slightly (e.g. "the invocation stops making further API calls" instead of "ends the process") or leaving as-is if you're confident in the inference — I don't think this rises to blocking, since the row's core claim (the two-item terminator set) is solidly verified either way.

Also purely stylistic: the immediately-preceding 0.12.1 entry links inline to the doc page it cites ([server-managed settings](...)); this new entry quotes goal.md verbatim but doesn't link it. Not a strict repo convention (most other entries don't link either), so optional.

No other issues found — plugin.json and the routines.md table diff are minimal and scoped correctly to the stated fix.

[claude]

Claude finished @kyle-sexton's task in 43s —— View job

---

Security Review — No Issues Found

• Gather context (gh pr diff / git diff origin/main...HEAD)
• Review plugins/autonomy/.claude-plugin/plugin.json diff
• Review plugins/autonomy/CHANGELOG.md diff
• Review plugins/autonomy/reference/routines.md diff
• Post findings

This PR touches only three files, all documentation/metadata, with no security-relevant surface:

• plugins/autonomy/.claude-plugin/plugin.json — a version bump string (0.12.1 → 0.12.2). No schema, permission, or dependency-pin changes.
• plugins/autonomy/CHANGELOG.md — a new changelog entry (prose only).
• plugins/autonomy/reference/routines.md — one glossary-table cell edited to correct the goal terminator description (documentation prose, no code, no config, no command execution).

No GitHub Actions workflows, hooks, MCP configs, scripts, or executable code paths are touched. There's no injection surface (command/SQL/path/template), no credential/token handling, no permissions:/trust-boundary changes, and no dependency or action pins to assess. This diff carries no security risk.

Verdict: No security findings.

[github-actions]

Claude has reviewed this PR 1 time. The lane skips further automatic reviews after 5; deleting this comment resets the count.