Proposed edit to skills/productivity/grilling/SKILL.md and docs/productivity/grilling.md. Replacement text first; the reason after.
skills/productivity/grilling/SKILL.md
Add, after the paragraph beginning "Each round the user answers reshapes the tree":
If the user's AGENTS.md or CLAUDE.md asks for one question at a time, ask one question per round rather than the whole frontier: put a single question, wait for the answer, then put the next. The frontier still decides which question is askable; only the number you ask at once changes.
docs/productivity/grilling.md
Under Can I go back to one question at a time?, replace:
Add this to your global CLAUDE.md:
with:
Add this to your AGENTS.md, or to your global CLAUDE.md:
and add, after the code fence:
The skill reads that line itself, so it works wherever your agent keeps your instructions: Codex reads AGENTS.md, Claude Code reads either.
Under Where did /batch-grill-me go?, "the CLAUDE.md line above is the way back to one-at-a-time" becomes "the instruction line above is the way back to one-at-a-time".
The reason
Two things stop the documented opt-out from reaching the agent.
The rule lives only on the docs page. SKILL.md never mentions it, so whether the line takes effect depends on the model noticing an instruction the skill's own text does not acknowledge. The page calls the opt-out "supported rather than tolerated", which reads as a promise the skill does not carry.
The file named is the wrong one for Codex. Codex users keep their instructions in AGENTS.md, so the documented CLAUDE.md is not the file being read. The page is already inconsistent with itself on this: its answer about a model that starts building instead of asking says "a line in your own AGENTS.md or CLAUDE.md".
Nothing changes for anyone who has not opted out.
Branch
The change, with a changeset, is on a public branch: https://github.com/itsmylife44/skills/tree/fix/grilling-opt-out-in-skill
Opening this as an issue rather than a pull request, since pull requests are limited to collaborators.
Proposed edit to
skills/productivity/grilling/SKILL.mdanddocs/productivity/grilling.md. Replacement text first; the reason after.skills/productivity/grilling/SKILL.mdAdd, after the paragraph beginning "Each round the user answers reshapes the tree":
docs/productivity/grilling.mdUnder Can I go back to one question at a time?, replace:
with:
and add, after the code fence:
Under Where did
/batch-grill-mego?, "theCLAUDE.mdline above is the way back to one-at-a-time" becomes "the instruction line above is the way back to one-at-a-time".The reason
Two things stop the documented opt-out from reaching the agent.
The rule lives only on the docs page.
SKILL.mdnever mentions it, so whether the line takes effect depends on the model noticing an instruction the skill's own text does not acknowledge. The page calls the opt-out "supported rather than tolerated", which reads as a promise the skill does not carry.The file named is the wrong one for Codex. Codex users keep their instructions in
AGENTS.md, so the documentedCLAUDE.mdis not the file being read. The page is already inconsistent with itself on this: its answer about a model that starts building instead of asking says "a line in your ownAGENTS.mdorCLAUDE.md".Nothing changes for anyone who has not opted out.
Branch
The change, with a changeset, is on a public branch: https://github.com/itsmylife44/skills/tree/fix/grilling-opt-out-in-skill
Opening this as an issue rather than a pull request, since pull requests are limited to collaborators.