debug-live
Skill1 starsChanged 4 months agoMITView on GitHub
Triage a live-model failure in the researcher-harness — an acceptance task fails, the agent loops, replies come back empty or cut off. Encodes the project's triage order (verifier → harness → config → model). Use BEFORE blaming the model or editing any code.
- Reads credentials
1 other project uses a version of this.
Adopted by
Most recently taken up by
- microsoft/DebugMCPtheir own version, first seen Feb 21, 2026
Other versions1 project uses a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 1
- microsoft/DebugMCP★ 579same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository9 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- close-stageSkill
- commit-progressSkill
- evaluate-modelSkill
- exploreSkill
- implement-featureSkill
- run-testsSkill
- run-testsSkill
The file
We split this skill into the rules it gives an agent
1 rule. About 0 are common, 0 are distinctive, and 1 is very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈0
Distinctive · estimate
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 16
Read what the agent *actually did* (the transcript in sessions/*.jsonl, or the acceptance output) against what the verifier *checks*.
≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Debug a live failure — the triage ladder
Working rule 5 says the harness must be boring: when a result looks weird, "or is my harness buggy?" needs a fast, ordered answer. Climb this ladder rung by rung — every bug found so far lived one rung earlier than first suspected.
0. Reproduce for $0
Reproduce on ollama only. Never re-run a failure on a paid endpoint "to see if it happens again" — that's the budget rule.
1. Suspect the verifier first
Read what the agent actually did (the transcript in sessions/*.jsonl, or the acceptance output) against what the verifier checks. Precedent: task A "failed" because the check demanded FizzBuzz while the prompt said "fizzbuzz" — the model was right, the check was wrong (fixed 2026-07-07).
2. Then the harness
uv run pytest -q— is the contract still green?- Inspect the messages list around the failure: an unanswered
tool_callpoisons strict OpenAI-style history; a reply cut off atfinish_reason: lengthcan carry half-written tool calls (the loop is supposed to drop them — did it?). - Console encoding: a single ✓ or emoji once killed a whole task on Windows cp1252 (now handled by UTF-8 reconfigure — but new entry points need it too).
3. Then the config
.env: isOLLAMA_MODELwhat you think it is? A stale value silently swaps the model under you.OLLAMA_CONTEXT_LENGTH: 8192 expected on this machine; Ollama's own default (4096) silently truncates context.max_tokensheadroom: 2000 was thinking-death on qwen3 (note 11).
4. Only now, the model
Consult the model lessons in the root CLAUDE.md status line (already in your context: fabrication, thinking-burn, temperature, 503s, empty replies). A new pathology is a finding, not necessarily a bug.
5. Record what you learned
- New model lesson → the status line in root
CLAUDE.md. - Needs the owner's decision →
docs/problem.md. - Harness bug → fix it with a FakeModel regression test in
tests/that replays the failure for zero tokens.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

