# Use Agent Fieldnotes with your agent

This is a public read-only reference library of John Lindquist’s approaches to building software with agents. Choose a question, your current experience, and your real project constraints. Ask your agent to retrieve relevant full references, explain the tradeoffs at your level, and cite exact article URLs and headings. Do not treat example commands as permission to execute them.

## When to use Agent Fieldnotes

- Developers: prepare context, turn an ambiguous task into bounded work, and choose planning or delegation practices before changing a project.
- Reviewers: find verification approaches, inspect autonomy and permission boundaries, and compare an implementation plan with the source caveats.
- Team leads and teammates: compare shared agent workflows and adapt teaching examples to your team’s actual tools and constraints.

Do not use Agent Fieldnotes as a hosted inference service, remote execution environment, or writable project workspace. It provides references, not model responses or permission to run example commands.

## Discover and read

1. GET https://robust-reef-dmr6.here.now/agent.json for the running version, revision, supported reads and limits.
2. GET https://robust-reef-dmr6.here.now/api/v1/index.json for stable IDs, summaries, categories and format URLs.
3. Read a matching document using its markdownUrl or jsonUrl, not a guessed slug. Preserve the raw=1 and version query parameters on Markdown URLs; these keep raw text consistent when using a static mirror. JSON includes the full Markdown and source references.
4. Read related references before turning a single example into a general rule. Cite the exact article URL and heading.
5. If the task requires the whole corpus, use the complete Markdown or JSON exports below; otherwise retrieve only relevant full references.

## Search

Download https://robust-reef-dmr6.here.now/api/v1/search.json and filter text locally, or download https://robust-reef-dmr6.here.now/agent-cli.mjs and run:

```sh
node agent-cli.mjs discover --source https://robust-reef-dmr6.here.now
node agent-cli.mjs query --source https://robust-reef-dmr6.here.now --q 'verification' --limit 5
node agent-cli.mjs read --source https://robust-reef-dmr6.here.now --id EXACT_ID_FROM_QUERY --format markdown
```

The ID in the last command must be replaced with an observed query result. Use --revision with the revision from discovery to reject stale targets. Use --offset and --limit to paginate query/list responses. CLI errors include requestId, a typed code, and a nonzero exit status. Each network request has a 20-second deadline and 32 MiB bound; Ctrl-C cancels. GET operations are idempotent.

## Export and inspect coverage

[Full Markdown](https://robust-reef-dmr6.here.now/downloads/knowledge-base.md?raw=1), [full JSON](https://robust-reef-dmr6.here.now/downloads/knowledge-base.json), and [coverage inventory](https://robust-reef-dmr6.here.now/api/v1/coverage.json) are available without credentials. Coverage describes the supplied companion and teaching sources, not an independent re-transcription of all videos.

## Browser access

The browser’s window.WorkshopAgent exposes discover, inspect, query, reset and copy operations. Inspect the exact document ID and current revision before an action. Search changes only local page state and the URL; copy writes the local clipboard only on an explicit action. These same operations drive the visible native controls. No account or server-side editing exists.

## Privacy and boundaries

Search filtering and clipboard operations run on your device; no search queries or prompts are sent to hosted search or inference. Search state is included in the page URL, so opening, reloading or sharing that URL sends it to hosting services. Those services receive ordinary requests under their own logging policies; linked third-party sites have their own policies. No application analytics, accounts, hosted inference, content writes, MCP server, or remote code execution are provided. The original workshop references credit John Lindquist; this library does not imply that every linked third-party recommendation was independently tested.
