fix(unified): emit C3.x output for local sources (#363)

[yusufkaraaslan]

Fixes #363

Problem

The unified skill builder dropped all C3.x analysis output produced by local sources. _run_local_codebase_analysis correctly extracted test_examples, patterns, how_to_guides, config_patterns, architecture, etc., wrote them to cache (local_analysis_{idx}_{name}/), and stored them on scraped_data["local"] — but unified_skill_builder.py only consumed GitHub sources for C3.x reference generation and SKILL.md summarization.

Concretely, both code paths only looked at GitHub:

• _generate_references() at line 1095-1101 iterated scraped_data.get("github", []) and called _generate_c3_analysis_references per repo.
• _generate_skill_md() at line 884-896 read scraped_data.get("github", {}) and passed only the first GitHub source's c3_analysis to _format_c3_summary_section.

So a unified config with a local source and extract_tests=true would happily run the test-example extractor (122 files / 292 examples in the issue), then quietly throw the result away.

The bug is broader than the issue title suggests — it's not just test_examples, it's the entire C3.x payload from local sources.

Solution

Refactor and extend in unified_skill_builder.py:

1. Refactor _generate_c3_analysis_references to delegate to a new shared helper _write_codebase_analysis_references(c3_data, source_id, github_data=None) that does the actual work. The GitHub path is now a thin lookup wrapper.
2. Add _generate_local_codebase_analysis_references that walks scraped_data["local"], skips sources with no C3 fields, and calls the shared writer with a sanitized source ID. Wired into _generate_references next to the GitHub loop.
3. Add _sanitize_source_id to make local IDs filesystem-safe (slashes, spaces, special chars → _; empty → "local").
4. Add _collect_c3_payloads that gathers C3 data from both GitHub and local sources into a flat list.
5. Rewrite _format_c3_summary_section to accept either a single dict (legacy) or a list of payloads, aggregating counts (test examples, design patterns, how-to guides, config files, security alerts) across all sources.

Testing

7 new regression tests in tests/test_unified.py:

• test_local_source_test_examples_become_references_363 — the headline issue: local source with test_examples produces references/codebase_analysis/<sanitized-id>/examples/test_examples.json and an ARCHITECTURE.md mentioning the example count.
• test_local_source_skill_md_includes_summary_363 — SKILL.md gets the C3 summary section when only local has data.
• test_format_c3_summary_aggregates_across_sources_363 — counts sum across multiple payloads.
• test_format_c3_summary_backward_compat_dict_arg_363 — single-dict input still works.
• test_collect_c3_payloads_combines_github_and_local_363 — combined harvesting from both source types.
• test_local_source_without_c3_data_skipped_363 — empty local sources don't create empty reference dirs.
• test_sanitize_source_id_363 — ID normalization edge cases.

All 31 tests/test_unified.py tests pass; 93 unified-related tests across test_unified.py + test_c3_integration.py + test_unified_scraper_orchestration.py + test_unified_analyzer.py all pass. Lint + format clean.

🤖 Generated with Claude Code

[codecov]

Codecov Report

❌ Patch coverage is 93.00000% with 7 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
src/skill_seekers/cli/unified_skill_builder.py	93.00%	7 Missing ⚠️

📢 Thoughts on this report? Let us know!