working-with-hal-apis

The workflow has two phases:

1. Explore — discover the API’s capabilities by following links and reading relation docs
2. Distill — export a recorded traversal as a deterministic path spec that can run without inference

<quick_start> Start a session at the API root, then explore from there:

npx hal-walk start -s session.json https://agent-wiki.mikekelly321.workers.dev/
npx hal-walk position -s session.json           # see current position + available links
npx hal-walk describe -s session.json wiki:v1:pages  # read the relation docs (markdown)
npx hal-walk follow -s session.json wiki:v1:pages    # follow the link

Every command except describe outputs JSON to stdout. describe outputs the raw markdown relation documentation. </quick_start>

<essential_principles> Follow your nose. Never construct URLs manually. Every action starts from the links available at your current position. If you need to reach a resource, navigate there through links.

Read before you act. When you encounter an unfamiliar link relation, use npx hal-walk describe to fetch its documentation. The docs tell you the HTTP method, required input schema, and expected response. This is how you figure out what to send — not by guessing.

CURIEs are namespaced relations. Links like wiki:v1:pages are CURIEs (Compact URIs). The prefix wiki: expands to a documentation URL via a template declared at the API root (e.g. wiki:v1:pages expands to .../rels/v1:pages). npx hal-walk describe handles this expansion automatically.

The session is a graph. Every follow records a position (node) and transition (edge). You can jump back to any previous position with goto without re-fetching. The graph can be rendered as a Mermaid diagram or exported as a path spec. </essential_principles>

start URL — Begin session, GET the root resource.

position — Show current position, response, and available links.

goto POSITION_ID — Jump to a previous position (local, no HTTP request).

describe RELATION — Fetch relation documentation (outputs markdown, not JSON).

follow RELATION — Follow a link relation from the current position.

render — Generate Mermaid diagram of the session graph.

export — Export a deterministic path spec from the session graph.

session-viewer — Serve a web UI for inspecting the session graph. Opens on a free port.

<exploration_workflow> Step 1: Start at the root

npx hal-walk start -s session.json https://agent-wiki.mikekelly321.workers.dev/

Read the output. Look at the links array — these are the relations you can follow.

Step 2: Understand a relation

Pick a relation from the links. Fetch its documentation:

npx hal-walk describe -s session.json wiki:v1:create-page

The markdown doc tells you:

• The HTTP method (POST, PUT, DELETE, GET)
• The input schema (what fields to send, which are required)
• An example request
• What the response looks like

Step 3: Follow the relation

Use what you learned from the docs to construct the follow command:

# For GET links (no body needed):
npx hal-walk follow -s session.json wiki:v1:pages \
  --note "List all pages to find existing content"

# For POST/PUT links (provide --body and --body-schema from the docs you read):
npx hal-walk follow -s session.json wiki:v1:create-page \
  --body '{"title": "My Page", "body": "Content here", "tags": ["demo"]}' \
  --body-schema '{"type":"object","properties":{"title":{"type":"string","description":"Page title"},"body":{"type":"string","description":"Page content (markdown)"},"tags":{"type":"array","items":{"type":"string"},"description":"Optional tags"}},"required":["title","body"]}' \
  --note "Create a new wiki page about demo content"

# For templated links (provide variables):
npx hal-walk follow -s session.json wiki:v1:search \
  --uri-template-values '{"q": "protocols"}' \
  --note "Search for pages about protocols"

# For PUT (explicit method override):
npx hal-walk follow -s session.json wiki:v1:edit-page \
  --method PUT \
  --body '{"body": "Updated content", "editNote": "Fixed typo"}' \
  --note "Fix typo in page content"

Step 4: Repeat

Each response reveals new links. Read the output, describe unfamiliar relations, follow the ones relevant to your task. Navigate deeper into the API graph.

Step 5: Backtrack if needed

npx hal-walk goto -s session.json p2

Jump back to any recorded position to explore a different branch. </exploration_workflow>

npx hal-walk render -s session.json    # visualize what you traversed
npx hal-walk export -s session.json    # export as a path spec

The path spec is a JSON array of steps with exact relations, methods, body schemas, example data, and notes. It captures the full workflow you discovered — including the contracts you learned and why you took each step — so it can be replayed by a simple runner with no inference.

To export a specific sub-path:

npx hal-walk export -s session.json --from p1 --to p5

This uses BFS to find the shortest path between two positions in the session graph.

<method_inference> The follow command defaults to GET. When you provide --body, it defaults to POST. For other methods (PUT, DELETE), use --method explicitly.

How to determine the correct method:

1. Run npx hal-walk describe on the relation
2. Read the “Method” section of the markdown
3. Use --method if it’s not GET or POST

Common patterns:

• create-* relations → POST (inferred from --body)
• edit-* relations → PUT (use --method PUT)
• delete-* relations → DELETE (use --method DELETE)
• Everything else → GET (default) </method_inference>

<schema_and_notes> Capture your inference. When describe output includes a JSON Schema for the input, pass it via --body-schema to preserve the contract in the session. This captures what you learned — required vs optional fields, types, descriptions — so it’s preserved in exports. If you omit --body-schema, a basic schema is auto-inferred from the body data, but this loses optionality info and descriptions.

The body is validated against the schema before sending. If they don’t match, the command fails immediately — this catches bugs in your data construction.

Leave breadcrumbs. Always provide --note with a brief description of why you’re taking each step. These notes are recorded on transitions and appear in exported path specs and the session viewer. They make your traversal self-documenting — someone reading the export understands not just what happened but why. </schema_and_notes>

<success_criteria> Exploration is successful when:

• You navigated the API entirely through link-following (no hardcoded URLs)
• You used describe to understand unfamiliar relations before acting
• The session file contains a complete record of your traversal

Distillation is successful when:

• npx hal-walk render produces a valid Mermaid graph showing the traversal
• npx hal-walk export produces a path spec with concrete steps, relations, methods, body schemas, and notes
• The path spec could be executed by a runner without any LLM inference
• Each step that sends data has a bodySchema capturing the contract and a note explaining intent </success_criteria>