ワンクリックでManusで任意のスキルを実行

coworld-episode-artifacts

スター2

フォーク2

更新日2026年7月1日 21:53

Use to identify and download Coworld episode artifacts — replays, results, and per-agent logs — for completed episodes, or STREAMED live from a still-running experience request (--watch: each episode downloads as it finishes, no waiting for the batch). Triggers: 'download the replay/logs for episode X', 'pull the last N episodes <player> played', 'fetch artifacts for this experience request / pool / round', 'stream artifacts while the request runs', 'get the episode data for post-mortem'. Game-agnostic (Crewrift, amongthem, etc.).

インストール

Codex または Claude でインストールこの Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。

Manusで実行

ソース

Metta-AI

Metta-AI/coworld-crewrift

GitHub リポジトリを開く Creator のリポジトリを見る

ダウンロード

Manusで実行

Coworld Episode Artifacts

Find completed Coworld episodes and download everything the Observatory data API will hand back for each — into one self-contained directory per episode, ready to replay, diff, or post-mortem offline. Game-agnostic: works for any Coworld game (Crewrift, amongthem, …), not one specific player.

Announce at start: "Downloading Coworld episode artifacts. I'll resolve the episodes for your selection, then pull replay + results + per-agent logs per episode."

What you get, per episode

One directory per episode (<timestamp>_<short-id>/) containing:

episode.json — the source record (league episode detail, or the experience-request episode row: participants, scores, status, game_config).
results.json — scores / metrics / win / per-agent outcomes.
replay.json + replay.json.z — the game replay, decompressed (the directly-loadable binary form) plus the raw zlib blob.
logs/policy_agent_{N}.log — each agent's full per-tick stderr trace (the richest behavioural record).
artifacts/policy_artifact_{N}.zip — any player-uploaded telemetry/debug bundles (policy-scoped: only slots you own come back; e.g. crewborg's trace zip with telemetry.jsonl + manifest.json). This is where you verify crewborg's LLM/Bedrock — telemetry.jsonl carries domain.meeting_llm_decision vs _fallback (see Bedrock debugging) — and it is not subject to the hosted ~10k-line log cap, so prefer it over logs/.
error_info.json — only when the episode failed.

Plus a top-level index.json summarizing the run. Every artifact is best-effort: a missing replay or one missing log is logged and recorded, never aborts the run.

The model (read this before debugging a 404)

Everything keys off job_id, the universal artifact handle every episode carries. Artifacts come from /jobs/{job_id}/artifacts/{results,replay,error_info} and /jobs/{job_id}/policy-logs[/{idx}]. There are two disjoint episode populations — league/tournament episodes (discovered by policy) and experience-request episodes (discovered by request/pool/round) — but both expose the same job_id, so one code path serves both. Full route map, dead-ends, and the recurring client/server drift: references/endpoint-map.md. Read it whenever a route 4xxs; the published coworld client regularly ships behind the server.

Workflow

Resolve live IDs first — never reuse cached ones. League/division/round/ pool/policy-version IDs rotate. For a policy by name the script resolves versions for you. For a pool/round/division/experience-request, get the current ID from the relevant CLI (coworld pools|rounds|divisions|xp-request list) or the Observatory UI and pass it in.

Pick exactly one discovery mode and run the downloader. Auth comes from softmax login; run inside uv run from an environment with the Coworld SDK (coworld[auth]) installed so softmax is importable.

F=players/crewborg/skills/coworld-episode-artifacts/scripts/fetch_artifacts.py

# A policy's most recent league episodes (across all its versions):
uv run python "$F" --policy crewborg -n 10 --out /tmp/crewborg_eps
# All child episodes of one experience request:
uv run python "$F" --xreq xreq_... --out /tmp/xreq_eps
# …or STREAM them while the experience request is still running (exits when drained):
uv run python "$F" --xreq xreq_... --watch --out /tmp/xreq_eps
# Explicit experience-request episodes (repeatable):
uv run python "$F" --ereq ereq_aaa... --ereq ereq_bbb... --out /tmp/eps
# Everything in a pool / round / division:
uv run python "$F" --pool pool_... -n 50 --out /tmp/pool_eps
# Explicit league episode uuids:
uv run python "$F" --episode <uuid> --out /tmp/eps

Useful flags: -n/--num (cap for policy/xreq/pool/round/division modes; default 10, unlimited under --watch), --version N (with --policy), --no-replay / --no-results / --no-logs (skip a category), --force (re-download complete dirs), --server (override the API base); watch mode adds --interval (poll seconds, default 15) and --max-attempts (retry bound, default 3). Runs are idempotent — complete episode dirs are skipped.

Use the artifacts. How to read each — the viewer vs the version-matched expand_replay vs the policy logs, the .bitreplay format, the slot↔policy mapping, and the hosted-log cap — is in crewrift-replays.md. In short: results.json + the episode row drive scoring; logs/ and the artifacts/*.zip telemetry hold per-agent behaviour; replays open in the viewer (uv run coworld replay <coworld_id> <replay.json>) or expand_replay. Turn a whole batch into a report with the crewrift-survey skill.

Notes

--watch (with --xreq only) polls the request and downloads each episode as it turns terminal instead of requiring the batch to be done — the streaming half of the default eval flow (see coworld-experience-requests step 4). Resume-safe: completeness is judged from disk, so a killed watch just picks up where it left off; watch_state.json bounds retries (3) for episodes whose artifacts error.
For interactive one-off inspection of a single experience-request episode, the coworld CLI (coworld episodes|replays|episode-logs|episode-results) is fine — see references/endpoint-map.md. This script is for discovering across a policy's versions and bundling everything per episode in one pass, and it reads raw routes so it survives client/server skew.
For creating the experience requests whose episodes this skill then downloads, use the coworld-experience-requests skill.
If auth fails with an AttributeError on load_current_cogames_token, you're looking at an older tool — the current API is load_current_token(server=...).

このリポジトリの他の Skills

同じリポジトリ

coworld-experience-requests

Metta-AI/coworld-crewrift

Use to create and monitor Coworld experience requests — hosted batches of episodes you define (target, roster, roles, count) for evaluating a policy against a live field. Triggers: 'run crewborg vs the top opponents', 'make an experience request', 'request N hosted games', 'A/B a policy against the league', 'measure the imposter', 'set up an evaluation battery'. DEFAULT FLOW after create: launch the STREAMING pipeline in the background (step 4) — artifacts download and analysis prep overlap the still-running episodes; don't wait for the batch to drain.

2026-07-012

coworld-local-run

Metta-AI/coworld-crewrift

Use to run your own built player policy in a LOCAL Coworld episode and watch it — the Gate-1 smoke test ('did my change take, does it connect → play → exit cleanly?'). Triggers: 'smoke test the player locally', 'run a local game', 'run my policy locally and watch it', 'does the build still work', 'open the replay locally'. Game-agnostic; this is correctness only, NOT a competitive matchup (you can't run other users' policies locally).

2026-07-012

coworld-policy-lifecycle

Metta-AI/coworld-crewrift

Use to submit an ALREADY-UPLOADED policy version to a Coworld league and monitor whether it QUALIFIES, then competes (and becomes champion). Triggers: 'submit crewborg to the league', 'did it qualify', 'is it disqualified / champion', 'watch the qualifier', 'monitor standings after submitting'. SUBMIT is the public, effectively-irreversible, champion-making action — explicit human go-ahead only. (Building + uploading a version is the separate `build & upload` skill.)

2026-07-012

crewrift-ab

Metta-AI/coworld-crewrift

Use to decide whether a crewborg change ACTUALLY helped — A/B the candidate against the baseline head-to-head, fresh, right now, against the same field. Triggers: 'did my change help', 'compare v71 vs v70', 'A/B test crewborg', 'is the candidate better', 'did fixing the imposter regress the crew'. Runs two MATCHED fresh experience requests, diffs role-split metrics with significance (compare.py), renders a comparison report, and guides a qualitative side-by-side. It's also the 'designed run' that crewrift-experiment hands off to.

2026-07-012

crewrift-event-warehouse

Metta-AI/coworld-crewrift

Use to build and query the Crewrift event warehouse — a policy-indexed DuckDB/Parquet dataset of per-tick gameplay events (kills, follows, proximity, votes, tasks, chat) over many episodes — for deep, mechanistic, cross-episode questions about a policy's behaviour. Triggers: 'how often is crewborg near a crewmate without killing', 'who trails crewborg', 'where does it abandon tasks', 'does it vote real imposters', 'build the warehouse for this round', 'query the events', 'stream an experience request into a warehouse'. For a FRESH experience request, stream_eval.py builds the warehouse WHILE the episodes run (the default; builds are incremental). This is the deep-dig mainstay after crewrift-survey flags something.

2026-07-012

crewrift-survey

Metta-AI/coworld-crewrift

Use to turn a SET of Crewrift episodes (an experience request, a policy's recent league games, a tournament batch) into a fast, polished HTML survey: a per-policy stats table, a policy×policy win heat map, and a short list of interesting episodes (with replay links). Triggers: 'survey these episodes', 'how is crewborg doing in this batch', 'who beats whom', 'give me the report on this XP request', 'which games should I watch'. Fast + lightweight (results.json + episode.json only). Pair with coworld-episode-artifacts (to pull the episodes); the DEEP per-episode dissection is a separate skill.

2026-07-012

name

coworld-episode-artifacts

description

Coworld Episode Artifacts

Announce at start: "Downloading Coworld episode artifacts. I'll resolve the episodes for your selection, then pull replay + results + per-agent logs per episode."

What you get, per episode

One directory per episode (<timestamp>_<short-id>/) containing:

episode.json — the source record (league episode detail, or the experience-request episode row: participants, scores, status, game_config).
results.json — scores / metrics / win / per-agent outcomes.
replay.json + replay.json.z — the game replay, decompressed (the directly-loadable binary form) plus the raw zlib blob.
logs/policy_agent_{N}.log — each agent's full per-tick stderr trace (the richest behavioural record).
artifacts/policy_artifact_{N}.zip — any player-uploaded telemetry/debug bundles (policy-scoped: only slots you own come back; e.g. crewborg's trace zip with telemetry.jsonl + manifest.json). This is where you verify crewborg's LLM/Bedrock — telemetry.jsonl carries domain.meeting_llm_decision vs _fallback (see Bedrock debugging) — and it is not subject to the hosted ~10k-line log cap, so prefer it over logs/.
error_info.json — only when the episode failed.

Plus a top-level index.json summarizing the run. Every artifact is best-effort: a missing replay or one missing log is logged and recorded, never aborts the run.

The model (read this before debugging a 404)

Workflow

Resolve live IDs first — never reuse cached ones. League/division/round/ pool/policy-version IDs rotate. For a policy by name the script resolves versions for you. For a pool/round/division/experience-request, get the current ID from the relevant CLI (coworld pools|rounds|divisions|xp-request list) or the Observatory UI and pass it in.

F=players/crewborg/skills/coworld-episode-artifacts/scripts/fetch_artifacts.py

# A policy's most recent league episodes (across all its versions):
uv run python "$F" --policy crewborg -n 10 --out /tmp/crewborg_eps
# All child episodes of one experience request:
uv run python "$F" --xreq xreq_... --out /tmp/xreq_eps
# …or STREAM them while the experience request is still running (exits when drained):
uv run python "$F" --xreq xreq_... --watch --out /tmp/xreq_eps
# Explicit experience-request episodes (repeatable):
uv run python "$F" --ereq ereq_aaa... --ereq ereq_bbb... --out /tmp/eps
# Everything in a pool / round / division:
uv run python "$F" --pool pool_... -n 50 --out /tmp/pool_eps
# Explicit league episode uuids:
uv run python "$F" --episode <uuid> --out /tmp/eps

Use the artifacts. How to read each — the viewer vs the version-matched expand_replay vs the policy logs, the .bitreplay format, the slot↔policy mapping, and the hosted-log cap — is in crewrift-replays.md. In short: results.json + the episode row drive scoring; logs/ and the artifacts/*.zip telemetry hold per-agent behaviour; replays open in the viewer (uv run coworld replay <coworld_id> <replay.json>) or expand_replay. Turn a whole batch into a report with the crewrift-survey skill.

Notes

--watch (with --xreq only) polls the request and downloads each episode as it turns terminal instead of requiring the batch to be done — the streaming half of the default eval flow (see coworld-experience-requests step 4). Resume-safe: completeness is judged from disk, so a killed watch just picks up where it left off; watch_state.json bounds retries (3) for episodes whose artifacts error.
For interactive one-off inspection of a single experience-request episode, the coworld CLI (coworld episodes|replays|episode-logs|episode-results) is fine — see references/endpoint-map.md. This script is for discovering across a policy's versions and bundling everything per episode in one pass, and it reads raw routes so it survives client/server skew.
For creating the experience requests whose episodes this skill then downloads, use the coworld-experience-requests skill.
If auth fails with an AttributeError on load_current_cogames_token, you're looking at an older tool — the current API is load_current_token(server=...).