trajrl

The official CLI for TrajectoryRL — an open skill factory that leverages Bittensor's distributed compute and incentive layer with reinforcement learning to produce state-of-the-art agent skills.

One install gives any human or AI agent (Claude Code, Cursor, Codex, OpenClaw, Hermes, Manus, ...) access to every skill TrajectoryRL has shipped. Each skill is a self-contained SKILL.md that agents can discover and follow directly.

CLI output is JSON when piped, tables when interactive.

Install

npm install -g trajrl

Subnet queries

# Network health
trajrl subnet status

# Validator analysis — scores, weights, scenarios, leaderboard
trajrl subnet analyze 5FFApaS7...
trajrl subnet analyze --uid 5 --deep

# Download a miner's pack (SKILL.md and evaluation metadata)
trajrl subnet download --uid 63
trajrl subnet download HOTKEY PACK_HASH

# Recent submissions
trajrl subnet submissions
trajrl subnet submissions --failed

Eval logs

Miner and validator logs are uploaded per evaluation and publicly downloadable.

# List recent eval logs for a validator
trajrl subnet logs --validator HOTKEY --limit 20

# List logs for a specific miner
trajrl subnet logs --miner HOTKEY

# Show the contents of a specific eval (summary + per-criterion scores)
trajrl subnet logs --eval-id 20260329_1430_w42 --show

# Extract the full archive locally for deep inspection
trajrl subnet logs --eval-id 20260329_1430_w42 --dump-to ./debug/

A miner eval archive contains:

SKILL.md                                 # miner's product
JUDGE.md                                 # scoring rubric used
metadata.json                            # final_score, delta, episode qualities
world.json                               # scenario context + salt
episodes/episode_N/
  testee_transcript.txt                  # agent's Hermes session log
  judge_transcript.txt                   # judge agent's grading log
  evaluation.json                        # per-criterion scores + summary
  episode.json                           # fixtures + instruction

Use this to debug SKILL.md iteration, inspect agent behavior, or audit any miner's eval end to end.

JSON output

Piped output is JSON by default. Use jq to compose queries:

trajrl subnet status | jq '.validators.validators[].hotkey'
trajrl subnet submissions | jq '.submissions[] | select(.evalStatus == "failed")'
trajrl subnet logs --eval-id <id> | jq '.logs[0].gcsUrl'

Force JSON in a tty with --json / -j. Override the API base URL with --base-url or TRAJRL_BASE_URL.

Skills

The skill catalog in this repo teaches agents how to use the CLI:

• skills/subnet-analyze/SKILL.md — query subnet data with trajrl subnet

Links

• Subnet repo: https://github.com/trajectoryRL/trajectoryRL — incentive mechanism, evaluation framework, Season 1 spec
• Bench: https://github.com/trajectoryRL/trajrl-bench — three-container eval sandbox (sandbox + testee + judge)
• Website: https://trajrl.com — leaderboard, live subnet data, dashboards
• Public API: PUBLIC_API.md — read-only, no auth, base URL https://trajrl.com