Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx agentmods add commands/evol-ai/skillcompass/eval-improvegit clone --depth 1 https://github.com/Evol-ai/SkillCompassWhat it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5 | $0.00000 | $0.02009 |
| Opus 5 | $0.00000 | $0.01005 |
| Sonnet 5 | $0.00000 | $0.00402 |
| Haiku 4.5 | $0.00000 | $0.00201 |
Grade A, and why
eval-improve scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 2d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 173 lines — stays where its author put it; the contents beside it link to each section on GitHub.
/eval-improve — Evaluate + Directed Improvement
Prerequisites
- Recommended model: Claude Opus 4.6 (
claude-opus-4-6). Directed improvement requires understanding complex rubric feedback and generating precise, targeted edits. Weaker models may produce unfocused rewrites that fail to address the weakest dimension or introduce regressions in other dimensions.
Interaction Mode
If --internal is passed, skip all interactive prompts — return results silently. This flag is set when eval-improve is called by other commands (e.g., from setup or skill-inbox action queue).
If --ci is passed, skip all interactive prompts and output pure JSON.
All user-facing text follows the global locale rule from SKILL.md.
Arguments
<path>(required): Path to the SKILL.md file to improve.--dimension D{N}(optional): Override which dimension to target. If omitted, targets the weakest.--feedback <path>(optional): Passed through to eval-skill.
Phase 1: Evaluate
Execute the full /eval-skill evaluation flow on the target skill. Use the Read tool to load {baseDir}/commands/eval-skill.md and follow its steps 1-18 with --scope full. Capture the complete JSON evaluation result.
Do NOT duplicate eval-skill logic here. Delegate entirely.
Phase 2: Snapshot
Use the Read tool to load {baseDir}/shared/version-management.md. Follow the snapshot procedure:
- Read or create the manifest for this skill.
- Compute the content hash of the current SKILL.md.
- Save a snapshot copy to
.skill-compass/{skill-name}/snapshots/{current-version}.mdusing the Write tool. - Create a transient self-write lock: use the Write tool to create
.skill-compass/.write-lockwith content{ "until": {unix_timestamp_now + 5} }. This prevents SkillCompass's own PostToolUse hooks from re-triggering during the confirmed improvement write.
Phase 3: Diagnose
From the evaluation result:
- If verdict is FAIL due to security gate: always target D3 first, regardless of
--dimensionflag. Security issues must be resolved before other improvements. - If
--dimensionwas specified: use that dimension as the target. - Otherwise: use the
weakest_dimensionfield from the eval result.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 2d ago First seen · 173 lines · 0 tokens per session scan A 2810760d7d84
eval-improve is a command published in the GitHub repository Evol-ai/SkillCompass (215 stars, last pushed 4mo ago), licensed MIT. It costs nothing until one of its globs matches a file; then it loads 2,009 tokens. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other commands, from other repositories
story-long-scan
长篇网文扫榜。分析起点、番茄、晋江等平台排行数据，提炼市场趋势。.
audit_docs
Audit all project metadata files for stale counts, version mismatches, broken references, and missing entries. Reports discrepancies without auto-fixing.
merge_session
Merge session branch(es) into main via rebase + fast-forward push. Use from inside a session worktree pane to land your work, or with --all to batch merge all sessions from the main repo.
cherry_pick_pr
Cherry-pick one or more commits onto a new branch from main and open a PR. Useful when a commit landed on the wrong branch or you want to split a multi-commit branch into separate PRs.
cti-report
Render case deliverables — relationship graph (PNG/SVG/Mermaid) and a polished PDF/DOCX assessment. Usage: /cti-report [--graph|--pdf].
reskin
Extract a measured design signature (type ramp, accent + its budget, grid unit, radius, layout) from a reference image or site, write it to signature.json, and drive the build to match it — "steal this vibe" as a spec, not pixels.