AGENT SKILL

compost

Run long or repetitive commands (builds, pytest/ctest, SLURM jobs, solver logs) through a clustering summarizer instead of reading raw output; re-runs show only what changed. Use for output likely over ~100 lines or fix-test-fix loops.

License 许可
MIT
最近更新
2026-11-13
Stars
★ 5
Forks
1

01 前言 Front Matter

name
compost
description
Run long or repetitive commands (builds, pytest/ctest, SLURM jobs, solver logs) through a clustering summarizer instead of reading raw output; re-runs show only what changed. Use for output likely over ~100 lines or fix-test-fix loops.
allowed-tools
Bash(${CLAUDE_SKILL_DIR}/bin/compost *), Bash("${CLAUDE_SKILL_DIR}/bin/compost" *)

02 资源文件 23

03 SKILL.md 正文

compost

!"${CLAUDE_SKILL_DIR}/bin/compost" mode --banner

Agents that do not expand the line above (e.g. Codex): run <this-skill-dir>/bin/compost mode --banner first and follow it.

Offline, deterministic: output is parsed and clustered by script (no LLM); the full raw output is always kept on disk. Skip it for short output -- just read that directly.

compost is not on PATH: always use the full <this-skill-dir>/bin/compost path (or python3 <this-skill-dir>/scripts/compost.py if executables are not permitted).

  1. Run: <this-skill-dir>/bin/compost run -- <command...> (pipes/redirects need bash -c '...'). Prints the root error (first error in user code), secondary failure groups with counts, warnings, numerical summary, and a raw_output_id.
  2. Read the summary, not the raw log.
  3. Re-run the same command the same way: only new/resolved errors and count deltas are printed ("No structural changes." if none).
  4. Only if the summary is insufficient, use the printed retrieval_handles (get <id> --lines A:B, event <id> --event N, grep <id> '<pattern>', show <id> --json). They never re-run the command; they read the already-stored raw output.
  5. Existing log file: bin/compost ingest --file <path> (or --stdin).
  6. The plugin's Bash hook may deny a raw build/test command and name the compost command to run instead: run that. Need raw output (e.g. to show the user)? Append # raw to the command.

Profiles (auto-detected, --profile to override): pytest, ctest, gcc/clang, cmake, python_traceback, slurm, numerical_solver, generic.

Output fields, statuses, storage and retention: references/schema.md. Per-profile detection: references/profiles.md.

04 同仓库更多 Skills

oliverweissl
oliverweissl/context-garden

file-organizer

Reorganize and rename files into tidy folder structures. Use when asked to sort, organize, or restructure a messy directory of files.

★ 5
oliverweissl
oliverweissl/context-garden

bloated-skill

Safely delete data files (CSV, temp, stale files) from a repository with mandatory user confirmation before every deletion. Use when asked to delete, remove, or clean up data files.

★ 5
oliverweissl
oliverweissl/context-garden

bloated-skill

This skill helps you with various tasks related to code and data. It is designed to assist you in many situations. Use it whenever you need help with things.

★ 5
oliverweissl
oliverweissl/context-garden

weeder

Measure and reduce a Skill's token footprint (always-loaded description + on-trigger SKILL.md body) without regressing routing accuracy or losing instructions. Use when a SKILL.md feels oversized, repeats rules, or has a vague description that might mis-route.

★ 5
oliverweissl
oliverweissl/context-garden

seedbank

Maintain a small, generated AGENTS.md of repository facts (build commands, conventions, invariants, past mistakes) that were rediscovered often or were costly to get wrong. Use at task start to check cached facts, and at task end to log and promote what you had to dig for.

★ 5
oliverweissl
oliverweissl/context-garden

pruner

Pick the smallest set of file:line ranges relevant to a concrete task (bug, compiler error, failing test, change) in a large Python or C/C++ repository, instead of grep-exploring or reading whole files. Skip when the task or traceback already names the file.

★ 5