Open data
As of , Agentic Feats lists 280 sourced examples of AI agents at work; 147 link the artifact. Every card carries its source link and its own absolute URL.
- Feats
- 280
- Shown
- 14753%
- Agents
- 20
- Updated
- Oct 14
Files schema 1.0
| File | Path | Size |
|---|---|---|
| All receipts, JSON | receipts.json | 498 KB |
| All receipts, CSV | receipts.csv | 161 KB |
| JSON Schema | schema.json | 1.0 |
| Indexed cards, markdown | llms-full.txt | 327 KB |
| Site map for agents | llms.txt | |
| One agent | /data/agents/{slug}.json | |
| One industry | /data/agents-for/{slug}.json | |
| One job type or role | /data/categories/… · /data/roles/… |
Use and stability
Every card carries its absolute address in url, in the JSON and the CSV.
Each file states its schema_version. Within version 1, fields are only added, never renamed or removed, so code written today keeps working. A breaking change ships under /data/v2/, and the version 1 files stay where they are.
The per-hub files hold up to 50 cards each, newest first, so one fetch stays small. When a hub has more, hub.next links the next page.
Grades, outcomes, and how cards are chosen are defined on the methodology page.
Fields 28 per card
- id
- Stable card id (ULID).
- slug
- URL slug: the title plus the first agent, frozen at first publish.
- url
- Absolute URL of the card page.
- title
- What the job was, in two to six plain words.
- summary
- What the agent did, in one line of our own words.
- result
- The concrete outcome, in one line.
- how_setup
- The agent plus the tools, integrations, or MCP servers used. Null when the source does not say.
- prompt_template
- A reusable prompt in our own words with bracketed placeholders. Not the author's exact prompt.
- caveat
- A limit, a cost, or a manual step.
- outcome_status
- completed, partial, or failed-with-lesson.
- evidence_grade
- A: an artifact is linked. B: a specific number is stated. C: a claim only. See /methodology/.
- evidence_url
- Link to the artifact (grade A).
- evidence_note
- What the evidence is, in a few words.
- agents
- The agents used, from the controlled list.
- category
- The job type, from the controlled list.
- roles
- Zero to two job roles, from the controlled list.
- industry
- Industry slug, or null for personal or general use.
- source_url
- The original post, repo, or article.
- source_type
- x, hn, github, blog, news, youtube, reddit, community, or submission.
- author_handle
- The source author's handle as written there. Never a looked-up real name.
- author_url
- The author's profile URL when the source gives one.
- project_url
- Link to the thing that was built.
- source_date
- When the source was published (YYYY-MM-DD).
- additional_sources
- Other URLs about the same use case.
- submitted_by
- Handle of the person who submitted the source, when given.
- pick
- Editor's pick.
- published_at
- When the card was published on Agentic Feats (ISO 8601).
- updated_at
- When the card last changed on Agentic Feats (ISO 8601).