Skip to content

fix: page advisor lint results - #324

Open
ahfoysal wants to merge 1 commit into
supabase:mainfrom
ahfoysal:fix/advisors-pagination
Open

fix: page advisor lint results#324
ahfoysal wants to merge 1 commit into
supabase:mainfrom
ahfoysal:fix/advisors-pagination

Conversation

@ahfoysal

@ahfoysal ahfoysal commented Jul 7, 2026

Copy link
Copy Markdown

What

Adds optional limit and offset inputs to get_advisors so large advisor responses can be fetched in smaller chunks.

This keeps the existing behavior unchanged when pagination options are omitted, but when either option is provided the returned lints array is sliced and a small pagination object is included with total, offset, limit, and next_offset when another page exists.

Fixes #124.

Validation

  • CI=1 ./node_modules/.bin/vitest run --project unit src/server.test.ts from packages/mcp-server-supabase
  • ./node_modules/.bin/tsc --noEmit from packages/mcp-server-supabase
  • ./node_modules/.bin/biome check packages/mcp-server-supabase/src/tools/debugging-tools.ts packages/mcp-server-supabase/src/server.test.ts packages/mcp-server-supabase/test/mocks.ts
  • git diff --check

@ahfoysal
ahfoysal requested a review from a team as a code owner July 7, 2026 11:17
anp0429 pushed a commit to anp0429/agentboard that referenced this pull request Jul 22, 2026
…ml; abort on base-less worktree review that misses the target

Three defects surfaced by one gating run on supabase/mcp#324:

1. Environment failures printed `stderr or stdout`: a harmless npm
   warning on stderr masked the real error on stdout
   (ERR_PNPM_RECURSIVE_EXEC_FIRST_FAIL). New _proc_tail() reports every
   non-empty stream, labeled. Applied to install/build/smoke paths in
   both verifiers.

2. supabase/mcp dropped packageManager and pins pnpm in mise.toml
   ([tools] pnpm = "10"); detect_pnpm_version fell back to 9, running
   the repo under a pnpm its workspace config was never written for.
   Detection now reads mise.toml and .tool-versions when package.json
   has no pin. packageManager still wins when present.

3. --worktree with no --base diffs the working tree against HEAD; on a
   clean PR checkout that is a near-empty diff, and the review ran
   anyway on intent alone (37 chars gated as if they were the PR).
   Now aborts with the fix named when the diff does not touch --target.
anp0429 pushed a commit to anp0429/agentboard that referenced this pull request Jul 22, 2026
…ct; existing tests are ground truth for setup and result shape

Two lessons from gating supabase/mcp#324, one run apart:

1. Nine confirmed gaps in a single run all failed with the verbatim
   message 'Target cannot be null or undefined.' — every generated test
   unwrapped the tool response with the wrong shape. Real gaps fail in
   their own words; artifacts fail in unison. New deterministic pass
   (flag_systematic_artifacts): >= 3 confirmed gaps sharing a verbatim
   first-line failure get an artifact_note and a run-level SYSTEMATIC
   FAILURE banner. Statuses never change; no model in the loop.

2. The run before that, 12 tests broke on setup copied from stale
   harness_notes while the diff's own added test demonstrated the
   correct seeding one screen away. The reviewer prompt now names the
   priority: the existing tests file (and the diff's own tests above
   all) are authoritative for setup and result unwrapping; harness_notes
   are background, and when they conflict, the tests win.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

get_advisors tool returns responses exceeding token limits without pagination support

1 participant