Back to skill
Skillv0.6.4
ClawScan security
AgentMeter · ClawHub's context-aware review of the artifact, metadata, and declared behavior.
Scanner verdict
BenignMar 6, 2026, 6:42 PM
- Verdict
- benign
- Confidence
- high
- Model
- gpt-5-mini
- Summary
- The skill's code and instructions match its stated purpose (local parsing of Claude Code transcripts, installing a Stop hook, storing spend records, and optional upload to a hosted dashboard); nothing in the bundle appears to request unrelated secrets or perform hidden exfiltration by default.
- Guidance
- This skill appears to do what it claims, but review and consent to the local changes before installing: 1) Back up .claude/settings.json (the script will modify or create it to add a Stop hook). 2) Inspect the included scripts (meter-session-end.sh, meter-parse-claude-sessions.sh, meter.sh, meter-sync.sh) yourself — they read ~/.claude/projects transcripts and write ~/.agent-meter/spend.jsonl. 3) The optional meter-sync.sh will upload spend records to api.agentmeter.io using an API key you provide; only enter a key if you trust that endpoint. 4) Ensure jq (and curl for sync) are available. 5) If you want to be cautious, run meter.sh / meter-parse-claude-sessions.sh in dry-run modes or inspect the created ~/.agent-meter files before enabling sync.
Review Dimensions
- Purpose & Capability
- okName/description (API spend tracking for Claude sessions) align with the scripts: they scan ~/.claude/projects for transcripts, install a session-end hook, compute token costs, and write records to ~/.agent-meter/spend.jsonl. Required files accessed and operations are appropriate for the stated goal.
- Instruction Scope
- noteSKILL.md directs the agent to run the included meter.sh which installs a Stop hook (copies meter-session-end.sh -> .claude/hooks/) and backfills transcripts. Those actions modify .claude/settings.json and write to ~/.agent-meter; this is expected for hooking/tracking but is a permanent local change the user should consent to. The hook and backfill parse full session transcripts (including user content) — that's necessary for per-message token accounting.
- Install Mechanism
- okNo remote install/downloads or obscure URLs. The skill is instruction-only (no external installer). All code is included in the bundle and runs locally; no extract-from-URL or third-party package installs are present.
- Credentials
- okThe skill requests no environment variables or credentials. The optional sync feature asks the user interactively for a dashboard API key which is stored locally at ~/.agent-meter/sync.json (chmod 600). That credential is proportional to the optional uploading feature.
- Persistence & Privilege
- noteThe skill writes files to ~/.agent-meter, copies a hook into .claude/hooks/, and edits/creates .claude/settings.json to register the Stop hook. These are required for ongoing capture but do grant the skill persistent local presence and the ability to read session transcripts. always:false (not force-included) and model invocation is normal.