B1 (agent-waiting-on-permission): body said 'allow / ask / deny' — inverted
vs the permissions doc's deny → ask → allow. Fixed; fact-check row 3 was
already correct and now matches the body.
B2 (agent-run-without-asking): dropped the 'Bash(git commit *) allowed with
Bash(git push *) denied is a policy the agent cannot talk its way around'
example — the docs' own 'What a Bash rule doesn't match' table names
Bash(git push *) as not stopping git -C . push origin main and friends.
Sentence narrowed to the precedence claim, with a pointer to step 4's caveat.
N1 (agent-done-signal): the osascript desktop-notification example is in the
hooks guide (code.claude.com/docs/en/hooks-guide), not the hooks reference.
Body and fact-check row 4 fixed. Pair-1 citations standardised from
docs.anthropic.com to code.claude.com per the review's optional note.
Each page's status blockquote records the revision.
Sponsor fact-check card (FRE-199, interaction a7cba232) answered 2026-11-05
20:21Z: 'Approve both — move to approved/'. Status blockquotes updated to
record the verdict and the date; page bodies untouched.
Sponsor fact-check card (FRE-199, interaction 0101ee80) answered 2026-11-05
19:53Z: 'Approve both — move to approved/'. Status blockquotes updated to
record the verdict and the date; page bodies untouched.
One voice, one place, so agent-drafted assets stop inventing framing
per artifact. positioning.md carries the wedge, the category-noun rule
(freno-dev/AGENTS.md competitor ban), claims each backed by code, the
honest privacy scoping and the voice/banned-phrasings list. The tree
adds calendar.json, per-channel briefs, drafts → pending → approved →
posted, metrics and listening — with the hard rule up front: approved/
is the only thing a human ever publishes from, nothing posts itself.
Issue: FRE-17
