triage-fix
Skill2 starsChanged 43 days agoMITView on GitHub
Drives one deterministic repair cycle after the test suite has failed - runs the full suite, localises the suspect frame from the traceback, reads a bounded code window, asks the student for a single search/replace patch, applies it, and re-runs the full suite. Use it whenever a task has failing tests and nobody has yet turned the traceback into a concrete file/line plus an attempted edit.
The file
We split this skill into the rules it gives an agent
1 rule. About 0 are common, 1 is distinctive, and 0 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈1
Distinctive · estimate
This author's own choices. Worth a look.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 19
If nothing fails, the task is marked done immediately.
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 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
triage-fix
Turns a traceback into a location and a patch instead of a reading tour.
What it does
- Runs
pytestwith no selector solast_full_runis always a real full-suite result. If nothing fails, the task is marked done immediately. - Takes the first failure, drops test/conftest/site-packages frames, and keeps the
innermost remaining project frame as
suspect={file, line, function}. - Reads a bounded window (20 lines back, 45 forward) into
context_snippet. - Asks the student exactly one question — purpose
patch— with the test, error, traceback and that window, and demands a verbatim search/replace for that file. - Applies it through
edit_file(never undertests/) and re-runs the full suite.
Bounds
- At most 3 runs per task, then
ask-studenttakes over. - Two patches that cannot be applied set
triage_giveup, and the routine stops applying. - An already-seen
(file, search)pair is never re-applied.
When to use it
Right after start, on any task whose suite is red. Do not use it for semantic judgement —
that stays with the student; this routine only owns the sequence of steps around it.
How Manifest uses it
Grown by Manifest's growth loop in round 2 (teacher: deepseek-flash); never hand-written.
scripts/routine.py exports NAME, applies(state) (trigger) and run(state, tools, student).
Trigger: Runs immediately after start on any task that is not yet done, up to 3 times, and once more is disabled after two patches the student could not apply. It first runs the full suite (marking the task done if it is already green), then localises the top failure, and drives one diagnose-and-patch cycle.
Permissions (Warden manifest)
- read: **
- write: src/**
- commands: python -m pytest
- network: false
- verdict: ok
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.