Skip to content

cli.md

Latest commit

 

History

History
111 lines (76 loc) · 2.54 KB

cli.md

File metadata and controls

111 lines (76 loc) · 2.54 KB
diataxis_type reference
title CLI Reference
description Complete command-line interface reference for the /autoresearch command

CLI Reference

Command

/autoresearch [OPTIONS] <path>

Modes

Full Improvement Loop (default)

/autoresearch <skill-path>
/autoresearch <skill-path> --iterations <N>

Runs the complete improvement cycle on a skill.

Arguments:

  • <skill-path> — Path to the skill directory (must contain SKILL.md)

Options:

  • --iterations <N> — Maximum number of improvement iterations. Default: 5. Range: 1-10.

Requires:

  • evals/evals.json must exist in the skill directory
  • The skill-creator plugin must be installed

Creates:

  • {skill-name}-autoresearch/ workspace directory (sibling to skill directory)

Eval Doctor

/autoresearch --eval-doctor <skill-path>

Creates or improves evaluation cases for a skill. Does not run the improvement loop.

Arguments:

  • <skill-path> — Path to the skill directory

Options:

  • --eval-doctor — Activates eval doctor mode

Creates/Modifies:

  • {skill-path}/evals/evals.json
  • {skill-path}/evals/trigger-eval.json (optional)

Report

/autoresearch --report <workspace-path>

Generates a convergence report from an existing autoresearch workspace.

Arguments:

  • <workspace-path> — Path to the autoresearch workspace (the {skill-name}-autoresearch/ directory)

Options:

  • --report — Activates report mode

Requires:

  • results.tsv must exist in the workspace
  • At least v0/ snapshot must exist

Examples

# Run with defaults (5 iterations)
/autoresearch ./skills/my-skill

# Run with more iterations
/autoresearch ./skills/my-skill --iterations 10

# Create evals for a bare skill
/autoresearch --eval-doctor ./skills/my-skill

# Review previous run results
/autoresearch --report ./skills/my-skill-autoresearch

Exit Conditions

The improvement loop exits when:

Condition Trigger
Perfect score best_score >= 1.0
Stuck Last 3 consecutive actions are reverted
Max iterations Iteration count reaches --iterations value

Environment

  • CLAUDE_PLUGIN_ROOT — Set automatically by Claude Code. Points to the plugin installation directory.
  • Python 3.8+ required for scripts (snapshot, scoring, reporting)
  • skill-creator plugin must be installed for grading

Related