llms.txt@site git:20261021.47019c0A

Immutable. This exact content never changes and is served at /api/v1/blob/f951b27753517eae.

# harness-ledger

> harness-ledger works with Claude Code, Codex and Cursor: it reads the session transcripts your agent already saves, maps the project's harness (instructions, skills, subagents, commands, hooks, settings) and suggests evidence-based changes with their costs.

Numbers (time, tokens, money) come from a local script, never from the model. Transcripts stay on the machine; links that leave it carry only shapes, names and counts. Install the skill with `npx skills add guilhermebkel/harness-ledger -g` (it asks which agents to install for: Claude Code, Codex, Cursor), then ask the agent to run `audit-harness` (`/audit-harness` in Claude Code).

## Docs

- [README](https://github.com/guilhermebkel/harness-ledger#readme): what it does, install, report format
- [Releases](https://github.com/guilhermebkel/harness-ledger/releases): versions and notes
- [Security](https://github.com/guilhermebkel/harness-ledger/blob/master/SECURITY.md): what counts as a vulnerability

## Optional

- [Contributing](https://github.com/guilhermebkel/harness-ledger/blob/master/CONTRIBUTING.md)
- [Issues](https://github.com/guilhermebkel/harness-ledger/issues)