docs: mark candidates A, D and E as shipped in the adoption survey
Problem
The adoption survey (docs/design/competitive-landscape-adoption.md) still lists three candidates as unbuilt that are in the code. A session working from it scoped diff-seeded impact, index seeding and git-aware staleness as new work and only found they existed after reading the changelog.
What changed
Rows A, D and E in §2 and items 1, 3 and 4 in §3 now say what shipped and what is still open. The status line notes the 2026-10-11 check.
| Candidate | Shipped | Still open |
|---|---|---|
| A. Git-diff-seeded impact | 2026-09-25, #125: explore answers "my changes", "this branch" and main..HEAD from the diff | No eval task class measures it |
| D. Index export and import | 2026-09-26, #147: codegraph init in a new worktree copies a sibling's index and re-reads the differing files | Seeding without someone running init (a policy choice) and export or import beyond sibling worktrees; the 9-of-11 count is not re-measured since #147 |
| E. Git-aware staleness | indexed_at_commit stamp and a git diff from it (#1829, #1878), a catch-up sync before the first query, a stale-file banner, opt-in post-commit, post-merge and post-checkout hooks | No catch-up cap found in the code; nothing measured |
Verification
- Each claim was read from the code and the changelog on
fork/consolidatedat ba3bce97:src/mcp/explore-changes.ts,src/sync/worktree-seed.ts,src/extraction/index.ts(beginGitIndexState,canTrustGitFastPath),src/sync/git-hooks.ts,src/mcp/tools.ts(formatStaleBanner). npm run check:agent-docspasses and the document's relative links resolve.- No eval task covers diffs:
.claude/skills/agent-eval/corpus.jsonholds only discovery questions, and nothing underscripts/agent-eval/mentions "my changes", "this branch" ormain..HEAD. Only a unit test (__tests__/explore-changes.test.ts) covers the feature. - Not checked: whether a catch-up cap exists outside the files above.
Summary by cubic
Marks candidates A (git-diff-seeded impact), D (index export/import), and E (git-aware staleness) in the adoption survey as shipped after a 2026-10-11 code check, so a future session doesn't re-scope finished work. Each table row and shortlist item now says what shipped and what's still open: A lacks a diff-change eval task, D still needs seeding without someone running codegraph init and export beyond sibling worktrees, and E bounds the first-query catch-up wait at 3 seconds but has no catch-up work cap. Claims were read from the code and changelog on fork/consolidated, and npm run check:agent-docs passes.
Summary by CodeRabbit
- Documentation
- Updated the competitive landscape assessment to mark diff-based impact analysis and sibling-worktree index copying as shipped, and commit-based staleness detection as mostly shipped.
- Clarified that the first query waits up to three seconds by default for index catch-up, stale results are flagged, and sync hooks are available.
- Documented remaining limitations, including the absence of a catch-up work cap, broader export/import support, and evaluation measurements for diff-based questions.