stuck-work-escalation

Before starting, read ~/.kmgraph/rules.md — Plan Protocol § Stuck-Work Escalation for thresholds, scope rules, and exit paths.

When This Applies

Trigger when work has a definable success criterion and has resisted resolution. Applies to:

• Bugs
• Failing tests (flaky vs. genuinely broken diagnosis)
• Blocked plan tasks
• Integration or API failures
• Performance targets that can't be hit
• Issues you can't reproduce

Does not apply to exploratory or iterative work (prompt tuning, CSS iteration, prototype spiking).

Workflow

At 3 Attempts (or 30 min) — powerful-tier Gate

Tier note: This skill declares required_tier: powerful-tier. It does not collapse to a lower tier — if powerful-tier is unavailable, the skill halts with an explicit error rather than running on a lower-capability model.

1. Create meta-issue (if not already open): /kmgraph:meta-issue "[Problem Title]"

2. Compile attempt log — read all prior attempts from meta-issue or conversation. For each attempt, confirm a distinct hypothesis was tested (not a retry of the same fix).

3. powerful-tier diagnosis — provide powerful-tier with:

   • Problem statement and success criterion
   • All attempts with hypotheses and results
   • Current root-cause-evolution.md if meta-issue exists powerful-tier reviews and provides fresh diagnosis and next hypothesis.

4. Log next attempt with powerful-tier-recommended hypothesis: /kmgraph:meta-issue --log-attempt NNN "[hypothesis]"

5. powerful-tier involvement continues — powerful-tier reviews each subsequent attempt result before the next attempt begins. Maximum 3 powerful-tier rounds before forcing exit-path analysis (step below).

At 5 Attempts — Exit-Path Analysis

Present the following to the user and require a decision:

Question	Answer
Would a brainstorming session reframe this?	Yes / No
Is there a known workaround?	[describe or No]
Is this required functionality or nice-to-have?	Required / Nice-to-have
Will the system break without it?	Yes / No / Partially
Will users be impacted?	Yes / No / [describe]
Can this be deferred to a future release?	Yes / No

Then select a mandatory exit path:

• Continue — only if powerful-tier identifies a genuinely new hypothesis not yet tried
• Defer — file KG issue, document blocker, move on
• Workaround — ship degraded version with limitation documented
• Descope — remove requirement; surface to user for confirmation
• Rescope — reframe the problem with a new success criterion
• User decision required — stop, present full context, wait

Present the exit-path analysis and selected path following plan approval rules. Wait for explicit approval before proceeding.

Counter Reset

If root cause genuinely shifts (new diagnosis invalidates prior attempts), reset the attempt counter. Log the reset in analysis/root-cause-evolution.md with reasoning.