Pod

Available as Markdown and JSON. Pod is also available over MCP.

Reported issues for Ryan Lindsey

Pod holds 8 of 8 GitHub reports that passed its relevance review. This can include external user reports, maintainer-confirmed bugs, and concrete feature gaps. Treat them as evidence to inspect, not a count of distinct defects.

Back to Ryan Lindsey.

Most discussed

fix(fit): the deferred run is cancelled at 30 seconds and the report never completes

Epic #270 moved the analyze_fit call into ctx.waitUntil so /fit/run could redirect immediately. ctx.waitUntil is capped at 30 seconds and the run takes 78.

Found by the whole-epic review after all seven children merged. Nothing in the spec, the epic, or any comment in the change names this budget, and no test can see it.

The measurement against the limit

analyze_fit was measured at 78,222 ms on 2026-09-18, trace 77c557ab4623b2fa059f29c7f75053b2. That number is quoted in…

Read the thread · 2026-09-22 · closed · 2 comments

fix(mcp): the one catch-all error log drops the error message, in the form measured not to work

workers/mcp/src/define.ts:308 logs the error as a second argument, which is the one form this codebase established does not work.

console.error(`mcp/${spec.surface}: ${spec.auditName} failed`, error);

Measured 2026-09-17 and recorded in workers/mcp/src/fit-start.ts: Cloudflare Worker observability renders console.error(msg, err) as the message followed by the stack and drops err.message entirely. That measurement is why completeRun interpolates instead:


[Read the thread](https://github.com/ryanlindsey/ryanlindsey.me/issues/354) · 2026-09-22 · closed · 1 comment

### fix(evals): payloadOf takes the first data line without matching on id, on the one SSE reader with live callers

`payloadOf` is the SSE reader with live production callers, and it carries only half the lesson that cost `/fit` eighteen days.

```ts
// workers/mcp/src/evals-client.ts
const data = text.split('\n').find((line) => line.startsWith('data:'));

It takes the first data: line in the stream and never matches on the JSON-RPC id. src/lib/fit/client.ts used to do exactly this, and the comment on its replacement records what it cost: fit_reports held zero rows from #37 until the fix,…

Read the thread · 2026-09-22 · closed · 1 comment

Spec: a /connect page for setting up the MCP server

Why

Every link on this site labelled MCP sends a person somewhere they cannot use. AGENT_LINKS in src/lib/nav.ts points the footer's FOR AGENTS column and the mobile overlay's strip at MCP_ENDPOINT, and /chat points its PREFER A PROTOCOL? rail and its <noscript> note at the same address. Measured 2026-09-24: a browser GET https://mcp.ryanlindsey.me/mcp with Accept: text/html answers 405 with application/json. The address is right for a client and a dead end for a reader.…

Read the thread · 2026-09-24 · closed · 0 comments

The fit workflow test reads the instance before /fit/start has created it

Measured 2026-09-23. tests/fit-workflow.test.ts › "the deferred run is a workflow instance named by the permalink id" failed once in CI and passed on the same code everywhere else.

The failure

Release PR #381, checks run 35883324048, at 777c78d:

FAIL tests/fit-workflow.test.ts > the deferred run is a workflow instance named by the permalink id
Error: instance.not_found
 ❯…

[Read the thread](https://github.com/ryanlindsey/ryanlindsey.me/issues/384) · 2026-09-23 · closed · 0 comments

### Stop `scripts/token.mjs` defaulting `--scopes` to the whole set

`scripts/token.mjs:286` defaults `--scopes` to every scope there is:

```js
const scopes = String(arg('scopes', SCOPES.join(',')))

So a mint that forgets the flag hands its audience fit,profile,documents,narrative,evals,authoring. Both withheld scopes are in that list, and both of them now open something.

Found during review of ryanlindsey/ryanlindsey.me#266, which is what made the second one true. Not fixed there: scripts/token.mjs is outside that issue's file list, and a mint…

Read the thread · 2026-09-18 · closed · 0 comments

bug(mcp): rate limiting does not enforce in production — 140 calls/second pass a 60/minute bucket

The RATE_LIMITER binding does not limit anything in production. Both MCP origins served 140 requests in 1 second against a documented 60/60s bucket, with zero refusals, and mcp_tool_calls has never recorded a single rate_limited outcome.

03 §3 requires every public tool call to be rate-limited. The audit half of that requirement is working (see below); the limiting half is not.

Evidence

140 parallel POSTs, get_contact, burst completed in 1s:
  mcp.ryanlindsey.me -> 0/140…

[Read the thread](https://github.com/ryanlindsey/ryanlindsey.me/issues/29) · 2026-09-08 · closed · 0 comments

### bug(mcp): ryanlindsey.me/mcp loops on its own origin — 7 of 8 tools fail with 522

Since #27 deployed, **`https://ryanlindsey.me/mcp` — the endpoint 03 §1 names as primary — fails for seven of the eight tools and for `resources/list`.** The vanity host `https://mcp.ryanlindsey.me/mcp` is unaffected and fully working.

## Symptom

| Call | `mcp.ryanlindsey.me` | `ryanlindsey.me` |
|---|---|---|
| `initialize` | ok | ok |
| `get_contact` | ok | ok |
| `request_private_access` | ok | ok |
| `get_resume` | ok | **`The résumé could not be read from the site right now.`** |
|…

[Read the thread](https://github.com/ryanlindsey/ryanlindsey.me/issues/28) · 2026-09-08 · closed · 0 comments

## Most recent

The remaining reports are on [the project's issue tracker](https://github.com/ryanlindsey/ryanlindsey.me/issues).