feat: repo-first evidence pipeline (v0.10.0)

[brianruggieri]

Summary

Implements the 6-decision repo-first evidence pipeline redesign from the 2026-03-30 grill session:

• D1 — RepoProject schema: Replaces ProjectSummary with RepoProject as canonical project source in MergedEvidenceProfile.projects. Projects now come from scanned repos via merge_triad(), not sessions.
• D2 — Commit evidence extraction: Two-stage pipeline — heuristic pre-filter (commit_filter.py) classifies commits into tiers and drops noise, then Claude extracts pithy highlight quotes with skill tags (commit_highlighter.py). Heuristic fallback when Claude is unavailable.
• D3 — Repo link wiring: repo_url and commit_url flow through select_projects() and select_evidence_highlights() to Hugo front matter.
• D4 — Three-tier skill resolution: Abstract skills (agentic-workflows, system-design, etc.) now resolve via mechanical match → commit tags (stub) → repo signal inference (static rules + Claude).
• D5 — Quantitative fallback: format_skill_repo_fallback() generates hiring-manager-aligned text from repo stats (repo count, timeline, test coverage, frameworks, CI) when no commit highlights exist.
• D6 — Session dormancy: export_fit_assessment() no longer reads candidate_profile.json. Session scanning infrastructure preserved but dormant.

New modules

• src/claude_candidate/commit_filter.py — heuristic pre-filter with tier classification
• src/claude_candidate/commit_highlighter.py — Claude-powered highlight extraction with fallback

Stats

• 22 files changed, +3118/−255 lines
• 1544 tests pass (28 skipped slow), up from 1423

Test plan

• Fast suite green: .venv/bin/python -m pytest (1544 passed, 28 skipped)
• Phase 1 spec review: all 8 tasks verified spec compliant
• Slow suite: .venv/bin/python -m pytest --run-slow
• Benchmark: .venv/bin/python tests/golden_set/benchmark_accuracy.py

Review fixes (f316048)

• fit_exporter.py: repo_url falls back to public_repo_url for legacy ProjectSummary shapes
• commit_filter.py: removed unused field import
• commit_highlighter.py: catch ClaudeCLIError specifically before broad Exception
• test_commit_filter.py: removed unused pytest import
• test_repo_project_schema.py: removed unused pytest import
• fit_exporter.py L1098: evidence_highlights empty by design (D6 dormancy, D2 wiring in follow-up) — no change needed

[Copilot]

Pull request overview

Implements the v0.10.0 “repo-first evidence pipeline” redesign by shifting MergedEvidenceProfile.projects to a repo-derived RepoProject, adding commit evidence extraction primitives (filter + Claude/heuristic highlighter), and updating export/merge paths and tests to reflect session dormancy.

Changes:

• Introduce RepoProject (replacing session ProjectSummary in merged profiles) and update merge/scoring call sites.
• Add commit evidence extraction pipeline: heuristic commit filtering + Claude-powered (with fallback) highlight extraction.
• Update fit export + tests for repo-first project shape, abstract-skill resolution scaffolding, and v0.10.0 version bumps.

Reviewed changes

Copilot reviewed 22 out of 22 changed files in this pull request and generated 6 comments.

Show a summary per file

File	Description
tests/test_repo_scanner.py	Adds coverage for raw commit fetching and optional highlight extraction in local repo scans.
tests/test_repo_project_schema.py	Adds tests for RepoProject schema and from_repo_evidence() factory behavior.
tests/test_repo_profile.py	Adds tests for CommitHighlight schema round-tripping and RepoEvidence defaults.
tests/test_merger.py	Updates merger tests to assert projects come from repo profiles (not sessions).
tests/test_merged_profile_projects.py	Verifies merged profiles accept RepoProject instances and serialize/deserialize.
tests/test_fit_exporter.py	Updates exporter tests for RepoProject-shaped projects, dormancy, repo_url/commit_url wiring, and abstract-skill helpers.
tests/test_commit_highlighter.py	Adds unit + slow integration coverage for commit highlight extraction and fallback.
tests/test_commit_filter.py	Adds tests for commit noise filtering, tiering, scoring, and slot budgeting.
src/claude_candidate/scoring/dimensions.py	Updates project-name usage to RepoProject.name.
src/claude_candidate/schemas/repo_profile.py	Adds RepoProject + CommitHighlight, and adds commit_highlights to RepoEvidence.
src/claude_candidate/schemas/merged_profile.py	Switches merged projects field type from ProjectSummary to RepoProject.
src/claude_candidate/schemas/init.py	Re-exports RepoProject from the schemas package.
src/claude_candidate/repo_scanner.py	Adds remote URL detection, optional highlight extraction, and raw commit fetching/parser helpers.
src/claude_candidate/merger.py	Populates merged projects from repo_profile.repos via RepoProject.from_repo_evidence().
src/claude_candidate/fit_exporter.py	Adds repo quantitative fallback formatter + abstract skill resolution helpers; updates project selection for RepoProject shape; removes candidate_profile dependency from export.
src/claude_candidate/extractors/signal_merger.py	Marks session-based project building as dormant with updated docstrings.
src/claude_candidate/commit_highlighter.py	New: Claude-powered commit highlight extraction with heuristic fallback.
src/claude_candidate/commit_filter.py	New: heuristic pre-filter with tiering/noise detection and scoring.
src/claude_candidate/cli.py	Updates export-fit command to no longer pass candidate_profile_path into exporter.
src/claude_candidate/init.py	Bumps package version to 0.10.0.
pyproject.toml	Bumps project version to 0.10.0.
extension/manifest.json	Bumps extension version to 0.10.0.

Comments suppressed due to low confidence (1)

src/claude_candidate/cli.py:338

• The export-fit command no longer checks that candidate_profile.json exists, but it still unconditionally reads it via CandidateProfile.from_json(candidate_path.read_text()). If the file is missing, this will raise a low-level exception/stack trace. Consider restoring an explicit existence check (or catching the error) to emit a clear CLI message and exit code.

	data_dir = Path.home() / ".claude-candidate"
	db_path = Path(db) if db else data_dir / "assessments.db"
	candidate_path = data_dir / "candidate_profile.json"

	# Validate paths
	if not db_path.exists():
		click.echo(f"Error: Database not found at {db_path}", err=True)
		raise SystemExit(1)

	# Load assessment from DB
	async def _load():
		store = AssessmentStore(db_path)
		await store.initialize()
		try:
			return await store.get_assessment(assessment_id)
		finally:
			await store.close()

	assessment = asyncio.run(_load())
	if not assessment:
		click.echo(f"Error: Assessment '{assessment_id}' not found.", err=True)
		raise SystemExit(1)

	# Build merged profile on the fly — no merged_profile.json needed
	cp = CandidateProfile.from_json(candidate_path.read_text())
	merged = _merge_profile(cp, quiet=True)