Index git history by time window and rank file ownership from authorship

[Phoenixrr2113]

What changed

• Persisted history window. Git history indexing no longer stops at a flat 200 commits. A new graph resolves its window once (365 days back, 10,000-commit initial-backfill ceiling), persists it on the Project node, and only ever widens: an earlier historySince or larger historyMaxCommits replays idempotently; narrower values never prune anything. Options flow through the MCP codebase reindex action, the raw trigger_reindex tool, POST /api/parse/project, and the CLI's --history-since / --history-max-commits flags, all with strict validation that now rejects impossible calendar timestamps (February 30 no longer silently becomes March 2).
• Two latent data-loss bugs fixed. An incremental backlog larger than the cap could advance the checkpoint past unprocessed commits, permanently skipping them; incremental sync now processes every commit after the checkpoint. A forced full reindex deleted File nodes and never rebuilt their history edges; it now replays the previously indexed window with no ceiling and restores the checkpoint (live probe: 23 edges before, 23 after, 24 after the next commit).
• New analyze action: ownership. Per-file contributor rankings from the author identity already indexed on every Commit node, with exact-pair grouping, share percentages, contributor truncation, an unknown-identity count, and honest caveats (inferred authorship, not CODEOWNERS; bots included; renames can split history). Available via MCP and GET /api/analysis/ownership. History coverage across hotspots, change coupling, and ownership now reports the persisted window, with historyWindowSize kept one release as a compatibility alias.
• Duplicate git syncs removed: MCP reindex and CLI extract no longer run a second sync with a different cap. A zero-commit repo now reports coherent empty coverage. Shared contracts live in packages/types (HistoryWindowOptions, HistoryCoverage, ownership result shapes).

Verification

Two fixers with disjoint ownership plus orchestrator-frozen types, red-first evidence throughout. Round-1 independent verification against real git fixtures found four blockers (the full-reindex replay cap, the calendar rollover, incoherent zero-commit coverage, and graph-layer limit clamping); all were remediated and round 2 probed the remediation deltas themselves (replay fallbacks, timezone-offset acceptance, strict-limit ripple): ACCEPT, zero blockers, with the reviewer's unmodified probes re-run at exit 0. Merged tree after pulling main: build 21/21 forced, tests 36/36 tasks, typecheck 20/20.

Deliberately not in this PR

• No dashboard UI for history options or ownership (dashboard keeps its path-only indexing flow).
• Ledger: built graph package has extensionless ESM imports blocking direct plain-Node import (pre-existing); symlinked project roots discover zero files (pre-existing); integration suites relocate long temp DB paths into ~/.codegraph/graphs/ (test isolation); one dashboard test fixture describes limit 100 where the live default is 50.

🤖 Generated with Claude Code

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
v0-landing-page-build	Ready	Preview, v0	Aug 24, 2026 12:16am