Diagnose
Portable equal-primary Claude and Codex agent harness for a governed agentic SDLC
npx -y skills add mblauberg/provenant --skill diagnoseAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
2 things to look at
- 27 days oldThe repository was created 27 days ago. New is not bad, but a brand new repository carrying a familiar-sounding name is the shape a typosquat arrives in, and there has been no time for anyone else to find a problem with it.
- 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.
What its author says it does
Copied from the file, not written here
Use to reproduce and find the root cause of broken behaviour, crashes, test failures, flakes, or performance regressions. Not for a known-cause repair or change review; use implement or code-review.
SKILL.md
2.4 KB, as published. Nobody here has run it
Diagnose
Build a mental model and check recorded constraints before probing. Work in order; record why any phase is skipped.
Iron law
NO FIX WITHOUT A REPRODUCTION AND A ROOT CAUSE FIRST
Diagnosis is read-only except temporary instrumentation. Permanent repair
requires an explicit user request or an authorised implement scope.
Workflow
- Feedback loop. Create the fastest sharp signal that reaches the reported failure: failing test, CLI/HTTP/browser fixture, replay, property loop, bisection, differential harness or structured user loop. Pin time/randomness and isolate mutable dependencies. For flakes, raise the reproduction rate. If no useful loop is possible, stop and request the missing environment or artifact; do not guess.
- Reproduce. Confirm repeated runs show the user's exact symptom, not a nearby failure. Capture the observation needed to prove later repair.
- Hypothesise. Rank 3–5 falsifiable causes before testing. Each needs a distinct prediction; inspect recent diffs, deploys and dependency changes.
- Instrument. Map one probe to one prediction and vary one thing at a time. Prefer debugger/breakpoints; tag targeted temporary logs uniquely. Performance work starts with baseline, profile and bisect.
- Handoff or fix. Without repair authority, stop with root cause, confidence, affected paths and the proposed regression seam. With authority, first turn the repro into a failing test at the real call-site seam, then make one root-cause fix and rerun the regression, original loop and suite. An absent correct seam is an architectural finding, not permission for a shallow test.
- Clean up. Remove only instrumentation/harnesses proven run-owned and covered by authority; preserve unknown, pre-existing and user-authored work. Record the winning hypothesis and prevention opportunity. Never bundle unrelated refactoring.
After three failed fix attempts, stop and question the architecture with the user. An enclosing lifecycle's stricter repair-cycle cap always wins. Detailed loop choices and cleanup checks: references/method.md.