Summary

Adds a read-only find_duplicate MCP tool that returns ranked duplicate candidates for an existing issue, gated behind a new opt-in duplicate_detection feature flag.

Why

Enables agents to answer "what likely duplicates this issue?" using GitHub's semantic-similarity endpoint, without adding schema cost to agents that haven't opted into duplicate detection. Semantic ranking stays server-side; the tool is a thin, read-only client.

Closes https://github.com/github/plan-track-agentic-toolkit/issues/690

What changed

• Add find_duplicate tool (Issues toolset, read-only) calling GET /repos/{owner}/{repo}/issues/{issue_number}/semantically_similar
• Add duplicate_detection feature flag to AllowedFeatureFlags (deliberately not in InsidersFeatureFlags so it's never implicitly enabled)
• Map confidence_threshold→threshold, perPage→per_page, page; omit optional params when unset so the API's high-precision defaults apply. No client-side bounds on confidence_threshold — the API owns the scale (observed scores exceed 1)
• Return a trimmed candidate — issue { number, title, state, url } (via the existing MinimalIssueRef) plus ranking metadata (score nullable, confidence, likely_duplicate) — rather than the full issue/repository payload
• Surface a clear error for legacy bare-issue responses instead of a success-shaped empty list
• Add unit tests, tool snapshot, and regenerate docs/feature-flags.md

MCP impact

• New tool added
  • Adds find_duplicate to the Issues toolset. It is non-default and only registered when the duplicate_detection feature flag is enabled via --features or X-MCP-Features.

Prompts tested (tool changes only)

Tested against a local build (--features duplicate_detection, repo-scoped PAT), via both Agent-mode prompts and direct JSON-RPC calls:

• Gating: find_duplicate is absent by default and present in tools/list only with the flag.
• "Using find_duplicate, find likely duplicates of issue 769 in github/issues-agentic-sandbox" → returned 585 ("Improve the onboarding flow for new users"), score ~1.93, confidence "high", likely_duplicate true.
• "Show the top 5 possible duplicates of issue 769 in that repo" (exercises perPage/confidence_threshold) → API returned the single genuine candidate; no error.
• Empty case — issue with no similar issues → [] (successful, not an error).
• Error case — a repo/issue where the endpoint isn't accessible → API 404 surfaced as an MCP tool error (isError: true), not a success-shaped empty result.

Security / limits

• Auth / permissions considered
  • Uses the caller's existing GitHub authentication; repository visibility and issue authorization are enforced by the API. The tool is read-only and mutates nothing.
• Data exposure, filtering, or token/size limits considered
  • Results are scoped to the source issue's repository (API-enforced), trimmed to a compact candidate shape to limit response size, and support pagination via perPage/page.

Tool renaming

• I am not renaming tools as part of this PR

Lint & tests

• Linted locally with ./script/lint
• Tested locally with ./script/test

Docs

• Updated (README / docs / examples)
  • Regenerated docs/feature-flags.md (documents the duplicate_detection opt-in and the read-only, single-repository find_duplicate tool). README needs no change — gated tools are intentionally excluded from its tool listing.