Filed by an autonomous agent under control-tower delegation, acting for the offline work-items authoring lane while blitzing PR #762 review threads.
Origin
Codex review finding on PR #762 (fix/540-work-items-shared-dogfood-skill), thread on plugins/work-items/reference/dogfood-filing.md line 74. A minimal in-doc guard landed on #762 (commit on that branch); this issue tracks the proper single-source fix, which reaches beyond that docs-only PR.
Repro / evidence (verified against source)
triage/SKILL.md "AI disclaimer" hardcodes triage-specific wording: > *This was generated by AI during triage.*. dogfood-filing.md ("Autonomous authorization and the AI disclaimer") routes every autonomous lane — work, scan-todos, source-control:babysit-prs, control-tower, enrichment — to prefix its filed items/comments with that same disclaimer. A non-triage lane reusing it verbatim stamps "during triage" on an item it filed during a work / TODO-scan / babysit cycle, mislabeling the provenance phase and weakening the audit trail.
Not defer-forbidden (why the SSOT fix is a follow-up)
The disclosure of AI authorship — the material transparency property — still holds; only the phase label is wrong. Audit-trail accuracy, not a safety-control failure. The #762 in-doc guard stops the doc from actively instructing the mislabel; this issue is the root fix.
Fix direction
Establish one lane-neutral AI-disclaimer SSOT (e.g. > *This was generated by AI.*, or a form parameterized by the acting lane) that every autonomous lane — triage included — points at, instead of triage owning triage-specific wording that other lanes inherit. Design call: where the neutral SSOT lives (a shared work-items reference vs. generalizing triage/SKILL.md's section) and whether the lane name is appended. Once it lands, dogfood-filing.md's guard note and triage's section both collapse to that single pointer.
Filed by an autonomous agent under control-tower delegation, acting for the offline
work-itemsauthoring lane while blitzing PR #762 review threads.Origin
Codex review finding on PR #762 (
fix/540-work-items-shared-dogfood-skill), thread onplugins/work-items/reference/dogfood-filing.mdline 74. A minimal in-doc guard landed on #762 (commit on that branch); this issue tracks the proper single-source fix, which reaches beyond that docs-only PR.Repro / evidence (verified against source)
triage/SKILL.md"AI disclaimer" hardcodes triage-specific wording:> *This was generated by AI during triage.*.dogfood-filing.md("Autonomous authorization and the AI disclaimer") routes every autonomous lane —work,scan-todos,source-control:babysit-prs, control-tower, enrichment — to prefix its filed items/comments with that same disclaimer. A non-triage lane reusing it verbatim stamps "during triage" on an item it filed during a work / TODO-scan / babysit cycle, mislabeling the provenance phase and weakening the audit trail.Not defer-forbidden (why the SSOT fix is a follow-up)
The disclosure of AI authorship — the material transparency property — still holds; only the phase label is wrong. Audit-trail accuracy, not a safety-control failure. The #762 in-doc guard stops the doc from actively instructing the mislabel; this issue is the root fix.
Fix direction
Establish one lane-neutral AI-disclaimer SSOT (e.g.
> *This was generated by AI.*, or a form parameterized by the acting lane) that every autonomous lane — triage included — points at, instead of triage owning triage-specific wording that other lanes inherit. Design call: where the neutral SSOT lives (a sharedwork-itemsreference vs. generalizingtriage/SKILL.md's section) and whether the lane name is appended. Once it lands,dogfood-filing.md's guard note and triage's section both collapse to that single pointer.