中文 技能评估系统
ReviewAudited by ClawScan on May 10, 2026.
Overview
This appears to be a local skill-evaluation tool, but its bundled reports contradict their own scoring and may falsely approve low-scoring skills, so it should be reviewed before relying on it.
Treat this as a local evaluator rather than a fully polished approval authority. Before installing or relying on it, verify the Python files and paths, run it first without --improve, and manually check generated scores because the bundled reports show contradictory approval labels.
Findings (5)
Artifact-based informational review of SKILL.md, metadata, install specs, static scan signals, and capability signals. ClawScan does not execute the skill or run runtime probes.
You could be told a skill is approved even when the same report says it did not meet the approval threshold.
The report says the score is below the stated 70% approval threshold and labels it NEEDS_WORK, but still marks it APPROVED. For a skill evaluator, contradictory approval output can cause users or agents to over-trust failed evaluations.
Score: 64/100 🟠 NEEDS_WORK ... STATUS: ✅ APPROVED (score >= 70%)
Do not rely on the approval line alone; verify the scoring logic and review generated reports manually before using this to approve or publish skills.
Using the improvement mode may alter or recommend changes to the skill being evaluated.
The skill advertises an auto-improvement mode and includes it in the documented full evaluation flow. This is aligned with a skill-improvement tool, but it may change or influence target skill files depending on implementation.
提供 --improve 选项自动改进 ... python3 .../evaluator.py <skill-path> --verbose --improve
Run without --improve first, keep backups or version control for evaluated skills, and review any generated changes before applying them.
Documented commands may fail or may not run the files the documentation expects.
The manifest provides eval-skill.py at the root and does not list scripts/eval-skill.py or references/rubric.md. The registry also declares no required binaries even though the instructions use python3. These are provenance/packaging inconsistencies, not evidence of malicious behavior.
`eval-skill.py` ... 捆绑在 skills/axioma-skill-evaluator/scripts/ ... `references/rubric.md`
Confirm the actual installed file layout and dependencies before running the commands.
Installing the skill alone does not run code, but using it as documented will execute local Python scripts against a chosen skill directory.
The skill is described as instruction-only in the install section but its normal use requires running bundled Python scripts. Local command execution is expected for this evaluator, but users should be aware of it.
python3 evaluator.py <skill-path> --verbose ... python3 eval-skill.py <skill-path> --verbose
Run the scripts only from the installed skill directory, on skill paths you intend to evaluate, and inspect the scripts if your environment is sensitive.
Old local path information and prior skill-evaluation context may appear in the skill package or confuse later evaluations.
The package includes persisted reports from a prior environment, including local absolute paths and skill names. This is not credential exposure, but it is extra historical context that could be reused or shown during future evaluations.
Path: /media/ezekiel/Morgana/skills/AUTO_RESEARCH
Remove bundled historical reports if they are not needed, or ensure they are clearly marked as examples.