A community catalog of autonomous agents and agent bundles certified by passing TraceCore deterministic episode runs gated by GitHub Actions.
This repository currently targets TraceCore 1.1.2 as the public certification baseline.
awesome-certified-agents is an "awesome list" with a twist: every entry is a verified case study, not just a link. To appear in this catalog, an agent must pass tracecore run --strict-spec in public CI — producing an immutable, schema-validated artifact as evidence.
GitHub Actions is the public gate. No human approval alone can certify an agent; the workflow must pass first.
For public submissions, the supported baseline is tracecore>=1.1.2, and CI certifies entries against the pinned launch baseline tracecore==1.1.2 for deterministic review.
- Submit your first entry: docs/how_to_submit.md
- AI prompt templates for contributors: docs/ai_prompt_templates.md
- External contributor onboarding: docs/external_contributor_onboarding.md
- Certification contract: docs/certification.md
- Evidence expectations: docs/evidence_policy.md
- Maintainer release steps: docs/release_process.md
- Reviewer checklist: docs/reviewer_checklist.md
| Badge | Tier | Requirement |
|---|---|---|
 |
Certified | ≥1 frozen TraceCore task passing --strict-spec |
 |
Certified+ | ≥3 frozen tasks across ≥2 task suites |
 |
Provisional | Metadata merged; CI not yet passing (grace period) |
Featured examples for the public launch:
- Starter single-agent example:
test-agent - Reference agent example:
toy-agent - Reference Certified+ example:
tri-suite-agent - Reference API example:
rate-limit-agent - Reference operations example:
ops-triage-agent - Reference bundle example:
filesystem-duo-bundle
The catalog now includes certified examples across filesystem, API, operations, and a multi-suite Certified+ entry.
| Agent / Bundle | Description | Framework | Suites | Tasks | Maintainer | Certified |
|---|---|---|---|---|---|---|
| tri-suite-agent | Multi-task agent that covers filesystem, API, and operations tasks with a single deterministic task-aware policy. | pure-python |
filesystem, api, operations | 3 | @justindobbs | Certified+ |
| filesystem-duo-bundle | Two-stage filesystem bundle where a scout agent locates candidate files and an extractor agent commits the recovered API_KEY. | pure-python |
filesystem | 1 | @justindobbs | Certified |
| ops-triage-agent | Deterministic operations agent that triages incident artifacts and recovers ALERT_CODE. | pure-python |
operations | 1 | @justindobbs | Certified |
| rate-limit-agent | Deterministic API agent that retrieves ACCESS_TOKEN while honoring rate limits and transient failures. | pure-python |
api | 1 | @justindobbs | Certified |
| test-agent | Minimal filesystem explorer that proves the certification workflow is wired correctly. | pure-python |
filesystem | 1 | @justindobbs | Certified |
| toy-agent | Cautious filesystem explorer that extracts API_KEY via list_dir → read_file → extract_value → set_output with retry logic. | pure-python |
filesystem | 1 | @justindobbs | Certified |
Following the FastAPI guidance on creating virtual environments, isolate your TraceCore install before running any commands:
python -m venv .venv # Windows: use "py -3.12 -m venv .venv" if multiple Python versions
# Windows activation
.venv\Scripts\activate
# macOS / Linux activation
source .venv/bin/activate- Install TraceCore:
pip install "tracecore>=1.1.2" - Run your agent:
tracecore run --agent my_agent.py --task <task_ref> --seed <seed> --strict-spec - Ingest the latest matching artifact:
python scripts/ingest_run_artifact.py agents/<my-agent> --latest --agent agents/<my-agent>/agent.py --task <task_ref> - Create the entry files under
agents/<my-agent>/, includingagent.py,README.md,runs/<run_id>.json, andmetadata.yamlthat matches the committed artifact - Validate locally:
python scripts/validate_metadata.py agents/<my-agent>/metadata.yaml --show-tier - Open a PR — CI re-runs your agent live and a CODEOWNER reviews the artifacts
Full guide: docs/how_to_submit.md
Before opening a PR, make sure your entry is understandable to an external reviewer with no prior context beyond the submitted files and the TraceCore docs.
| Task | Suite | What the agent must do |
|---|---|---|
filesystem_hidden_config@1 |
filesystem | Discover API_KEY in a constrained filesystem |
rate_limited_api@1 |
api | Fetch ACCESS_TOKEN under rate limits and transient errors |
rate_limited_chain@1 |
api | Navigate a multi-step handshake + rate limit |
deterministic_rate_service@1 |
api | Parse payload templates + rate-limited service |
log_alert_triage@1 |
operations | Triage noisy logs, recover ALERT_CODE |
config_drift_remediation@1 |
operations | Detect config drift, emit remediation patch |
incident_recovery_chain@1 |
operations | Follow a recovery handoff chain, emit RECOVERY_TOKEN |
log_stream_monitor@1 |
operations | Poll paginated logs, detect CRITICAL, emit STREAM_CODE |
runbook_verifier@1 |
operations | Verify runbook phase order, emit RUNBOOK_CHECKSUM |
sandboxed_code_auditor@1 |
operations | Audit a sandbox runtime, emit `ISSUE_ID |
All tasks are deterministic given a fixed seed. See TraceCore SPEC_FREEZE.
Every certified run artifact includes:
| Field | Purpose |
|---|---|
spec_version |
Declares tracecore-spec-v1.0 compliance |
runtime_identity |
Name, version, git SHA of the runtime |
task_hash |
SHA-256 of the task harness files |
agent_ref |
Agent module path invoked |
artifact_hash |
Stable SHA-256 of the serialized artifact |
budgets |
Frozen max steps/tool-calls for the episode |
wall_clock_elapsed_s |
Total episode wall time in seconds |
Artifacts conform to artifact-schema-v1.0.json and are immutable once committed.
- Submit an agent: CONTRIBUTING.md
- Step-by-step guide: docs/how_to_submit.md
- AI prompt templates: docs/ai_prompt_templates.md
- External contributor onboarding: docs/external_contributor_onboarding.md
- Certification spec: docs/certification.md
- Evidence policy: docs/evidence_policy.md
- Reviewer checklist: docs/reviewer_checklist.md
- Badge embedding: docs/badges.md
- Release process: docs/release_process.md
- Release notes: CHANGELOG.md
- Nominate an agent: open an issue
[](https://github.com/awesome-certified-agents/awesome-certified-agents)MIT — see LICENSE.
Powered by TraceCore — deterministic episode runtime for autonomous agents.