Intent

Consolidate all prose vocabulary for worker agents on crewmate and crewmates across shared tracked material, eliminating the former distinction where crew meant the collective.
Preserve every compatibility identifier containing crew exactly, including config paths, script and skill names, CLI modes, account-pool names, shell functions and variables, metadata keys, status strings, and fixtures.
The resulting diff must be prose/comment-only with zero behavioral changes, AGENTS.md must state the single-word rule, and every remaining crew or crews occurrence must be attributable to a compatibility identifier.

What Changed

• Standardize worker-agent prose on crewmate and crewmates across shared documentation, skills, script comments, and unasserted test comments.
• Clarify worker-pool and account-selection wording while preserving compatibility identifiers containing crew unchanged.
• Preserve asserted diagnostics, grep targets, fixture strings, runtime strings, and identifiers byte-for-byte.
• Keep the update documentation/comment-only with no runtime behavior changes.

Audit note

This ambiguity is a real instruction-surface defect, not an implementation error.
Tonight's audit found that the crewmate-vocabulary rule did not distinguish ordinary prose from asserted compatibility strings, which is why a careful crewmate had to stop and ask twice.
The clarified split is semantic: prose comments about direct reports use crewmate; asserted or matched strings and identifiers remain compatibility surface.

Risk Assessment

Low: the complete branch diff is documentation/comment-only, and no runtime line, asserted string, metadata key, status text, fixture, or identifier changes.

Validation

• bash -n passed for bin/*.sh and bin/backends/*.sh.
• Pinned bin/fm-lint.sh passed.
• tests/fm-account-directory.test.sh passed.
• tests/fm-bootstrap.test.sh passed.
• The review fix also ran affected-file targeted checks; fm-account-routing.test.sh reproduced its pre-existing session-sync race only after all preceding cases passed, while fm-crew-state.test.sh passed.
• The local no-mistakes test phase was deliberately skipped because the suite-wide behavior glob can enter real-Herdr lifecycle tests and endanger the live fleet. No suite-wide local behavior run was allowed.

Pipeline

Updates from git push no-mistakes

• Intent: passed
• Rebase: passed
• Review: one ambiguity finding resolved with the firstmate-directed prose-versus-compatibility split; re-review passed
• Test: skipped by explicit fleet-safety direction
• Document: passed
• Lint: passed
• Push: passed