Examples — Rendered Session Audits
Two complete walkthroughs of the report that the audit skill produces. Every value in them
is made up: hosts use example.test / example.com, tokens appear as <token> with a fake
8-hex fingerprint, and no real session is described.
To audit the real session, run
/session-doctor-dev:audit(addfullfor the raw DETAILS lines).
Available Scenarios
| Scenario | What it shows | Reference |
|---|---|---|
| A clean session audit | Nothing wrong: all tables filled, "No problems found" | references/scenarios/clean-session.md |
| A rejected token and a failed MCP server | A 401 on a token, a failed MCP server, and the Problems table with fixes | references/scenarios/rejected-token-failed-mcp.md |
How to read a report
| Section | Source in the data block | Meaning |
|---|---|---|
| Summary table | session header | where the session runs, model and effort, context size, MCP status |
| Skills | SKILLS INVOKED |
which skills ran, who invoked them, how they ended |
| HTTP | HTTP REQUESTS |
every request with its method, host, path and status codes |
| Tokens | TOKENS |
one row per token seen — variable name and fingerprint only, status from the codes received |
| Problems | findings + judgment | at most 8 rows, ordered red, yellow, low; each with evidence and a fix |
Rules the examples follow
- A token value is never printed — the Fingerprint column is a short hash, not a prefix of the secret.
- Token status comes only from codes the session already received; the audit makes no network request.
- An empty block is written as "none", not as an empty table.