work-items(triage): step-2 "wait for user direction" gate contradicts the skill's own autonomous/agent triage mode

[kyle-sexton]

Lane: triage · Category: behavioral

Triggering example

/work-items:triage run as an autonomous loop lane (standing rules authorize autonomous label/comment mutations, and the skill's own AI-disclaimer prefix is applied). Two parts of the triage SKILL.md directly contradict each other for this mode:

• Step 2 ("Recommend category + state") ends with: "Wait for the user's direction before mutating anything."
• The "AI disclaimer" section explicitly anticipates the opposite: "When creating comments or items during autonomous/agent triage sessions, prefix with…" — i.e. autonomous sessions that mutate without a human turn.

An operator following step 2 literally cannot triage autonomously; an operator triaging autonomously violates step 2. Nothing in the skill reconciles the two.

Observed vs expected

• Observed: the skill mandates a human-direction gate before any mutation (step 2), while simultaneously shipping autonomous/agent-mode affordances (the AI-disclaimer prefix) that presuppose unattended mutation. No documented branch selects between them.
• Expected: a documented autonomous/agent mode in which standing lane rules (or a userConfig/session flag) supply the "direction," lifting the step-2 wait-gate — so autonomous triage is not in permanent contradiction with step 2. Interactive single-issue triage keeps the wait-gate; autonomous lanes declare the gate satisfied by their standing rules.

Notes

Adjacent to #449 (no-local-checkout mode) but distinct: #449 is about the seam assuming a project root; this is about the human-in-the-loop gate assuming an interactive operator. Both surface when the skill runs unattended.

[kyle-sexton]

This was generated by AI during triage.

Triage: routed status: ready (ready-with-noted-decision)

Category: behavioral · Priority: priority: high (week directive: pipeline self-improvement) · Self-labeled severity: medium (documentation contradiction, no runtime failure) — noted for implementer sub-sort within high.

Verified

The contradiction reproduces in the current triage SKILL.md: step 2 ends with "Wait for the user's direction before mutating anything," while the AI disclaimer section presupposes "creating comments or items during autonomous/agent triage sessions." No branch reconciles them, so an operator cannot both triage autonomously and honor step 2.

Plan (behavioral contract)

Document two triage operating modes and scope the wait-gate to the interactive one:

• Interactive mode (single-issue, /work-items:triage <number> with a human operator): step 2's wait-gate stands unchanged — no mutation before the user directs.
• Autonomous / standing-lane mode (triage run under a standing loop lane whose rules authorize label/comment mutation): the standing lane rules constitute the "direction" step 2 waits for. The wait-gate is satisfied by that standing authorization; the skill proceeds to mutate, applying the AI-disclaimer prefix (already specified) to every comment/item it creates.
• Verification (step 3) is unchanged and still precedes any mutation in both modes — autonomy lifts the human-direction gate, never the verify-before-mutate discipline.
• Rewrite step 2's closing sentence to carry the mode scope explicitly so the sentence and the AI-disclaimer section no longer contradict.

Acceptance criteria

• A reader can determine from the skill alone, with no contradiction, whether a given invocation waits for human direction.
• The AI-disclaimer section and step 2 are mutually consistent.
• Autonomous mode still reproduces + verifies before mutating.

Decision defaulted: autonomous mode is selected by the standing lane rules themselves (the /loop prompt = the "direction"); no new userConfig/session flag is required — a flag is optional future work. — veto before merge

Dedup note

Issue #478 lists an "Autonomous-mode contract" bullet that is the same fix — that bullet defers to this issue; implement here, reference from #478.