Skill
audit
Check whether what I said in this session is backed by what I actually did — did the tests I reported as passing really run, was an expected value quietly…
@sjh9714View SkillDevelopment
Skill
/history
Add up every Claude Code session on this machine and report how often an agent said the tests passed and how often anything actually ran them. Reads the local transcripts under ~/.claude/projects and prints a coverage count plus any findings. Use when the user asks about their
Install
$ npx -y skills add sjh9714/red-handed --skill history --agent claude-codeHow it fires
How this skill gets triggered: by you, by Claude, or both.
- Fires itselfAuto-invocation. Claude auto-loads it when your prompt matches the work.Auto-invocation is when the right skill fires by itself at the right moment, driven by a FLOW.md router and a hook, instead of you invoking it by name. It is the difference between a skill being installed and a skill actually getting used.Read the full definition →
- You can call itInvoke it directly when you want it.
- Slash command
/history
Context preview
The summary Claude sees to decide when to auto-load this skill.
Add up every Claude Code session on this machine and report how often an agent said the tests passed and how often anything actually ran them. Reads the local transcripts under ~/.claude/projects and prints a coverage count plus any findings. Use when the user asks about their
SKILL.md
history.SKILL.mdname: history
description: "Add up every Claude Code session on this machine and report how often an agent said the tests passed and how often anything actually ran them. Reads the local transcripts under ~/.claude/projects and prints a coverage count plus any findings. Use when the user asks about their whole history rather than one session, or asks how often their agent has claimed something it did not verify."
allowed-tools:
- Bash(npx --yes @jinhyuk9714/red-handed@latest stats *)
- Bash(npx --yes @jinhyuk9714/red-handed@latest stats)
Audit every session on this machine
npx --yes @jinhyuk9714/red-handed@latest stats
Add `--lang ko` for a Korean report, `--since 30` to limit it to the last 30 days, `--json` to read the numbers rather than quote them, `--no-cache` to recompute from scratch.
The first run reads every transcript; later runs are near-instant because the results are cached in `~/.red-handed/cache.json` (mode 0600, on this machine only, relocatable with `RED_HANDED_HOME`).
What to report
The headline is the coverage count, which everyone gets whether or not anything was found:
184 sessions. Your agent said "tests pass" 84 times.
79 a test ran first
5 no test ran in that session at allQuote it as printed. Then note how many findings there were, if any.
Two things to keep straight when the user reads their own number:
- "No test ran in that session" is not proof of a lie. Verification this tool
cannot read — a project's own runner, a browser check — counts as verification it did not see. That is why those land at `SUSPICIOUS` rather than `CAUGHT`.
- A count of zero at the `CAUGHT` tier is a real result, not an empty one. The
tool is built to miss things rather than to accuse wrongly.
Read more
name: history description: "Add up every Claude Code session on this machine and report how often an agent said the tests passed and how often anything actually ran them. Reads the local transcripts under ~/.claude/projects and prints a coverage count plus any findings. Use when the user asks about their whole history rather than one session, or asks how often their agent has claimed something it did not verify." allowed-tools: - Bash(npx --yes @jinhyuk9714/red-handed@latest stats *) - Bash(npx --yes @jinhyuk9714/red-handed@latest stats)
Audit every session on this machine
npx --yes @jinhyuk9714/red-handed@latest stats
Add `--lang ko` for a Korean report, `--since 30` to limit it to the last 30 days, `--json` to read the numbers rather than quote them, `--no-cache` to recompute from scratch.
The first run reads every transcript; later runs are near-instant because the results are cached in `~/.red-handed/cache.json` (mode 0600, on this machine only, relocatable with `RED_HANDED_HOME`).
What to report
The headline is the coverage count, which everyone gets whether or not anything was found:
184 sessions. Your agent said "tests pass" 84 times.
79 a test ran first
5 no test ran in that session at allQuote it as printed. Then note how many findings there were, if any.
Two things to keep straight when the user reads their own number:
- "No test ran in that session" is not proof of a lie. Verification this tool
cannot read — a project's own runner, a browser check — counts as verification it did not see. That is why those land at `SUSPICIOUS` rather than `CAUGHT`.
- A count of zero at the `CAUGHT` tier is a real result, not an empty one. The
tool is built to miss things rather than to accuse wrongly.
Ships withred-handed
Audits what your Claude Code agent did against what it said — reads session logs and git, cites evidence, calls no model
Get the whole plugin
Stats
4
Stars
1
Forks
Maintained
Maintenance
TypeScript
Language
MIT
License
1mo ago
Last commit
1mo ago
Created
Repo: sjh9714/red-handed
Other skills on red-handed.
@sjh9714View Skill