Debrief a finished run

The agent that led the run debriefs once, after its final validation and before its final report. Subagents do not run this skill; fold the AutoformBot friction they reported to you into your own answer.

There is no autoform on PATH in a Lean project. Run it from <PROJECT> as uv run --project "<AUTOFORM_PLUGIN_ROOT>" autoform, where the root is the absolute plugin directory holding skills/, pyproject.toml, and autoform_cli/; give a subagent that prefix, with the root filled in.

Run autoform debrief orchestrator form <PROJECT> --skill setup|roadmap|formalize, naming the skill that ran (see the CLI reference). If it says orchestrator debriefs are disabled, stop: there is nothing to do.

Otherwise answer the form from your experience of the run and store the answer with autoform debrief orchestrator record <PROJECT> --skill <SKILL>, from --answer FILE or standard input. If record rejects it, correct and resubmit.

When the answer reports friction with AutoformBot, record writes a draft issue for facebookresearch/autoform-bot outside the working tree and prints its absolute path. The user does not see command output, so end your final report with the draft as a Markdown link to the whole path, exactly as printed and in angle brackets; the draft usually lives under .git, which editors hide: [Draft AutoformBot issue](</home/me/proj/.git/autoform/debriefs/issues/RUN_ID.md>). Say that nothing was sent and that the link at the bottom of the draft opens the issue form pre-filled with the recorded draft, so edits belong in the form; if record says the draft is too long for a link, say instead that the link fills in only the title. Do not paste that link. Never submit the issue yourself, with gh or otherwise: the user reviews it, edits it, and sends it. The draft is only as private as your answer, so leave project-specific content out of it.