Debug hard

Skill travelhawk/forgeshift/.claude/skills/debug-hard

Escalate a stubborn bug to the hard-bug debugging specialist with a structured handoff - it runs on the session model, so put the session on Fable 5 for the hardest ones when you have the quota. Use when a bug survived two fix attempts, reproduces intermittently, or looks impossible - instead of a third normal attempt.From its SKILL.md

Install

npx -y skills add travelhawk/forgeshift --skill debug-hard

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

2 stars2 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

SKILL.md

2.3 KB, as published. Nobody here has run it

/forge:debug-hard — Escalate, don't grind

The bug: "$ARGUMENTS". A third attempt with the same approach would fail like the first two — this skill changes the approach: full context, fresh eyes, the session's strongest model (switch the session to Fable 5 first if you have the quota).

1. Assemble the handoff (this is most of the value)

Write the dossier — from the session, git, and the user where needed:

Symptom: exact observed behavior vs expected, verbatim error output.
Reproduction: exact steps/commands and how reliably it fires (always? 1 in 10?). If unknown, say so — reproduction is then the first goal.
Timeline: when it started, what changed around then (git log the suspect window).
Failed attempts: each previous fix idea, what it changed, and how it failed. These are evidence about what the bug is NOT — the most valuable section.
Environment: where it happens (dev/CI/prod, OS, versions) and where it doesn't.

Don't include your current favorite theory as fact — label hypotheses as hypotheses.

2. Delegate

Hand the dossier to forge-temper. It works hypothesis-driven: reproduce → fault model → discriminating experiments → root-cause fix → regression test proven in both directions. Expect it to take its time — that's the point.

3. Verify & close

Confirm the regression test fails on pre-fix code and passes post-fix (the debugger proves this; spot-check the claim against its evidence).
For previously intermittent bugs: enough repeat runs that the old failure rate would have shown itself.
Record the root cause in one plain-language line in PROGRESS.md session log (trigger → mechanism → symptom). If it revealed a design weakness, propose the ADR.
Next → back to the lane you came from: /forge:feature//forge:build to resume the build, or /forge:ship if this bug was the last blocker. Name the one move.