feat: fix session_pair_harvester to use role/content format (#91)

- Harvester used old message fields (from/value) but Hermes sessions use role/content
- Import session_reader to normalize conversations properly
- Update extract function to operate on normalized role/content messages
- Change predecessor lookup from "human"/"gpt" to "user"/"assistant"
- Add comprehensive smoke tests (8 tests, all pass)
- Verify extraction from test_sessions: 11 pairs, avg ratio 8.13

docs/API.md

@@ -1,472 +0,0 @@
-# Compounding Intelligence — Scripts API Reference
-
-*Generated: 2026-04-26 11:02 UTC*
-
-This document auto-documents the public API surface of all scripts
-in `scripts/`. Each section covers one script: module purpose,
-public functions, and their signatures.
-
----
-
-## `scripts/api_doc_generator.py`
-
-API Doc Generator — Issue #98
-
-| Function | Signature | Description |
-|----------|-----------|-------------|
-| `extract_functions_from_ast` | `extract_functions_from_ast(tree, file_rel)` | Extract public function names, signatures, and first-line doc summaries. |
-| `parse_module` | `parse_module(filepath)` | Parse a Python file and return its module-level docstring and public functions. |
-| `scan_scripts_dir` | `scan_scripts_dir(scripts_dir)` | Scan all .py files in scripts/ and extract API info. |
-| `render_markdown` | `render_markdown(modules)` | Generate full docs/API.md content from the scanned modules. |
-| `render_json` | `render_json(modules)` | Emit machine-readable JSON version of the API reference. |
-| `main` | `main()` | - |
-
-## `scripts/automation_opportunity_finder.py`
-
-Automation Opportunity Finder — Scan fleet for manual processes that could be automated.
-
-| Function | Signature | Description |
-|----------|-----------|-------------|
-| `analyze_cron_jobs` | `analyze_cron_jobs(hermes_home)` | Analyze cron job definitions for automation gaps. |
-| `analyze_documents` | `analyze_documents(root_dirs)` | Scan documentation for manual step patterns. |
-| `analyze_scripts` | `analyze_scripts(root_dirs)` | Detect repeated command sequences in scripts. |
-| `analyze_session_transcripts` | `analyze_session_transcripts(session_dirs)` | Find repeated tool-call patterns in session transcripts. |
-| `analyze_shell_history` | `analyze_shell_history(root_dirs)` | Find repeated shell commands from history files. |
-| `deduplicate_proposals` | `deduplicate_proposals(proposals)` | Remove duplicate proposals based on title similarity. |
-| `rank_proposals` | `rank_proposals(proposals)` | Sort proposals by impact * confidence (highest first). |
-| `format_text_report` | `format_text_report(proposals)` | Format proposals as human-readable text. |
-| `main` | `main()` | - |
-
-## `scripts/bootstrapper.py`
-
-Bootstrapper — assemble pre-session context from knowledge store.
-
-| Function | Signature | Description |
-|----------|-----------|-------------|
-| `load_index` | `load_index(index_path)` | Load and validate the knowledge index. |
-| `filter_facts` | `filter_facts(facts, repo, agent, include_global)` | Filter facts by repo, agent, and global scope. |
-| `sort_facts` | `sort_facts(facts)` | Sort facts by: confidence (desc), then category priority, then fact text. |
-| `load_repo_knowledge` | `load_repo_knowledge(repo)` | Load per-repo knowledge markdown if it exists. |
-| `load_agent_knowledge` | `load_agent_knowledge(agent)` | Load per-agent knowledge markdown if it exists. |
-| `load_global_knowledge` | `load_global_knowledge()` | Load all global knowledge markdown files. |
-| `render_facts_section` | `render_facts_section(facts, category, label)` | Render a section of facts for a single category. |
-| `estimate_tokens` | `estimate_tokens(text)` | Rough token estimate. |
-| `truncate_to_tokens` | `truncate_to_tokens(text, max_tokens)` | Truncate text to approximately max_tokens, cutting at line boundaries. |
-| `build_bootstrap_context` | `build_bootstrap_context(repo, agent, include_global, max_tokens, index_path)` | Build the full bootstrap context block. |
-| `main` | `main()` | - |
-
-## `scripts/dead_code_detector.py`
-
-Dead Code Detector for Python Codebases
-
-| Function | Signature | Description |
-|----------|-----------|-------------|
-| `is_safe_unused` | `is_safe_unused(name, filepath)` | Check if an unused name is expected to be unused. |
-| `get_git_blame` | `get_git_blame(filepath, lineno)` | Get last author of a line via git blame. |
-| `analyze_file` | `analyze_file(filepath)` | Analyze a single Python file for dead code. |
-| `scan_repo` | `scan_repo(repo_path, exclude_patterns)` | Scan an entire repo for dead code. |
-| `main` | `main()` | - |
-
-## `scripts/dedup.py`
-
-dedup.py — Knowledge deduplication: content hash + semantic similarity.
-
-| Function | Signature | Description |
-|----------|-----------|-------------|
-| `normalize_text` | `normalize_text(text)` | Normalize text for hashing: lowercase, collapse whitespace, strip. |
-| `content_hash` | `content_hash(text)` | SHA256 hash of normalized text for exact dedup. |
-| `tokenize` | `tokenize(text)` | Simple tokenizer: lowercase words, 3+ chars. |
-| `token_similarity` | `token_similarity(a, b)` | Token-based Jaccard similarity (0.0-1.0). |
-| `quality_score` | `quality_score(fact)` | Compute quality score for merge ranking. |
-| `merge_facts` | `merge_facts(keep, drop)` | Merge two near-duplicate facts, keeping higher-quality fields. |
-| `dedup_facts` | `dedup_facts(facts, exact_threshold, near_threshold, dry_run)` | Deduplicate a list of knowledge facts. |
-| `dedup_index_file` | `dedup_index_file(input_path, output_path, near_threshold, dry_run)` | Deduplicate an index.json file. |
-| `generate_test_duplicates` | `generate_test_duplicates(n)` | Generate test facts with intentional duplicates for testing. |
-| `main` | `main()` | - |
-
-## `scripts/dependency_graph.py`
-
-Cross-Repo Dependency Graph Builder
-
-| Function | Signature | Description |
-|----------|-----------|-------------|
-| `normalize_repo_name` | `normalize_repo_name(name)` | Normalize a repo name for comparison. |
-| `scan_file_for_deps` | `scan_file_for_deps(filepath, content, own_repo)` | Scan a file's content for references to other repos. |
-| `scan_repo` | `scan_repo(repo_path, repo_name)` | Scan a repo directory for dependencies. |
-| `detect_cycles` | `detect_cycles(graph)` | Detect circular dependencies using DFS. |
-| `to_dot` | `to_dot(graph)` | Generate DOT format output. |
-| `to_mermaid` | `to_mermaid(graph)` | Generate Mermaid format output. |
-| `main` | `main()` | - |
-
-## `scripts/diff_analyzer.py`
-
-Diff Analyzer — Parse unified diffs and categorize every change.
-
-*(no public functions — script runs as `main()` only)*
-
-## `scripts/freshness.py`
-
-Knowledge Freshness Cron — Detect stale entries from code changes (Issue #200)
-
-| Function | Signature | Description |
-|----------|-----------|-------------|
-| `compute_file_hash` | `compute_file_hash(filepath)` | Compute SHA-256 hash of a file. Returns None if file doesn't exist. |
-| `get_git_file_changes` | `get_git_file_changes(repo_path, days)` | Get files changed in git in the last N days. |
-| `load_knowledge_entries` | `load_knowledge_entries(knowledge_dir)` | Load knowledge entries from YAML files in the knowledge directory. |
-| `check_freshness` | `check_freshness(knowledge_dir, repo_root, days)` | Check freshness of knowledge entries against recent code changes. |
-| `update_stale_hashes` | `update_stale_hashes(knowledge_dir, repo_root)` | Update hashes for stale entries. Returns count of updated entries. |
-| `format_report` | `format_report(result, max_items)` | Format freshness check results as a human-readable report. |
-| `main` | `main()` | - |
-
-## `scripts/gitea_issue_parser.py`
-
-Gitea Issue Body Parser — Extract structured data from markdown issue bodies.
-
-| Function | Signature | Description |
-|----------|-----------|-------------|
-| `parse_issue_body` | `parse_issue_body(body, title, labels)` | Parse a Gitea issue markdown body into structured JSON. |
-| `fetch_issue_from_url` | `fetch_issue_from_url(url)` | Fetch an issue from a Gitea API URL and parse it. |
-| `main` | `main()` | - |
-
-## `scripts/harvester.py`
-
-harvester.py — Extract durable knowledge from Hermes session transcripts.
-
-| Function | Signature | Description |
-|----------|-----------|-------------|
-| `find_api_key` | `find_api_key()` | Find API key from common locations. |
-| `load_extraction_prompt` | `load_extraction_prompt()` | Load the extraction prompt template. |
-| `call_llm` | `call_llm(prompt, transcript, api_base, api_key, model)` | Call the LLM API to extract knowledge from a transcript. |
-| `parse_extraction_response` | `parse_extraction_response(content)` | Parse the LLM response to extract knowledge items. |
-| `load_existing_knowledge` | `load_existing_knowledge(knowledge_dir)` | Load the existing knowledge index. |
-| `fact_fingerprint` | `fact_fingerprint(fact)` | Generate a deduplication fingerprint for a fact. |
-| `deduplicate` | `deduplicate(new_facts, existing, similarity_threshold)` | Remove duplicate facts from new_facts that already exist in the knowledge store. |
-| `validate_fact` | `validate_fact(fact)` | Validate a single knowledge item has required fields. |
-| `write_knowledge` | `write_knowledge(index, new_facts, knowledge_dir, source_session)` | Write new facts to the knowledge store. |
-| `harvest_session` | `harvest_session(session_path, knowledge_dir, api_base, api_key, model, dry_run, min_confidence)` | Harvest knowledge from a single session. |
-| `batch_harvest` | `batch_harvest(sessions_dir, knowledge_dir, api_base, api_key, model, since, limit, dry_run)` | Harvest knowledge from multiple sessions in batch. |
-| `main` | `main()` | - |
-
-## `scripts/improvement_proposals.py`
-
-Improvement Proposal Generator for compounding-intelligence.
-
-| Function | Signature | Description |
-|----------|-----------|-------------|
-| `analyze_sessions` | `analyze_sessions(sessions)` | Analyze session data to find waste patterns. |
-| `generate_proposals` | `generate_proposals(patterns, hourly_rate, implementation_overhead)` | Generate improvement proposals from waste patterns. |
-| `format_proposals_markdown` | `format_proposals_markdown(proposals, patterns, generated_at)` | Format proposals as a markdown document. |
-| `format_proposals_json` | `format_proposals_json(proposals)` | Format proposals as JSON. |
-| `main` | `main()` | - |
-
-## `scripts/knowledge_gap_identifier.py`
-
-Knowledge Gap Identifier — Pipeline 10.7
-
-*(no public functions — script runs as `main()` only)*
-
-## `scripts/knowledge_staleness_check.py`
-
-Knowledge Store Staleness Detector — Detect stale knowledge entries by comparing source file hashes.
-
-| Function | Signature | Description |
-|----------|-----------|-------------|
-| `compute_file_hash` | `compute_file_hash(filepath)` | Compute SHA-256 hash of a file. Returns None if file doesn't exist. |
-| `check_staleness` | `check_staleness(index_path, repo_root)` | Check all entries in knowledge index for staleness. |
-| `fix_hashes` | `fix_hashes(index_path, repo_root)` | Add hashes to entries missing them. Returns count of fixed entries. |
-| `main` | `main()` | - |
-
-## `scripts/perf_bottleneck_finder.py`
-
-Performance Bottleneck Finder — Identify slow tests, builds, and CI steps.
-
-| Function | Signature | Description |
-|----------|-----------|-------------|
-| `find_slow_tests_pytest` | `find_slow_tests_pytest(repo_path)` | Run pytest --durations and parse slow tests. |
-| `find_slow_tests_by_scan` | `find_slow_tests_by_scan(repo_path)` | Scan test files for patterns that indicate slow tests. |
-| `analyze_build_artifacts` | `analyze_build_artifacts(repo_path)` | Find large build artifacts that slow down builds. |
-| `analyze_makefile_targets` | `analyze_makefile_targets(repo_path)` | Analyze Makefile for potentially slow targets. |
-| `analyze_github_actions` | `analyze_github_actions(repo_path)` | Analyze GitHub Actions workflow files for inefficiencies. |
-| `analyze_gitea_ci` | `analyze_gitea_ci(repo_path)` | Analyze Gitea/Drone CI config files. |
-| `find_slow_imports` | `find_slow_imports(repo_path)` | Find Python files with heavy import chains. |
-| `severity_sort_key` | `severity_sort_key(b)` | Sort by severity then duration. |
-| `generate_report` | `generate_report(repo_path)` | Run all analyses and generate a performance report. |
-| `format_markdown` | `format_markdown(report)` | Format report as markdown. |
-| `main` | `main()` | - |
-
-## `scripts/priority_rebalancer.py`
-
-Priority Rebalancer — Re-evaluate issue priorities based on accumulated data.
-
-| Function | Signature | Description |
-|----------|-----------|-------------|
-| `collect_knowledge_signals` | `collect_knowledge_signals(knowledge_dir)` | Analyze knowledge store for coverage gaps and staleness. |
-| `collect_staleness_signals` | `collect_staleness_signals(scripts_dir, knowledge_dir)` | Run staleness checker if available. |
-| `collect_metrics_signals` | `collect_metrics_signals(metrics_dir)` | Analyze metrics directory for pipeline health. |
-| `extract_priority` | `extract_priority(labels)` | Extract priority level from issue labels. |
-| `compute_issue_score` | `compute_issue_score(issue, repo, signals, now)` | Compute priority score for a single issue. |
-| `generate_report` | `generate_report(scores, signals, org, repos_scanned)` | Generate the full priority report. |
-| `generate_markdown_report` | `generate_markdown_report(report)` | Generate human-readable markdown report. |
-| `main` | `main()` | - |
-
-## `scripts/refactoring_opportunity_finder.py`
-
-Finds refactoring opportunities in codebases
-
-| Function | Signature | Description |
-|----------|-----------|-------------|
-| `compute_file_complexity` | `compute_file_complexity(filepath)` | Compute cyclomatic complexity for a Python file. |
-| `calculate_refactoring_score` | `calculate_refactoring_score(metrics)` | Calculate a refactoring priority score (0-100) based on file metrics. |
-| `scan_directory` | `scan_directory(directory, extensions)` | Scan directory for source files. |
-| `generate_proposals` | `generate_proposals(directory, min_score)` | Generate refactoring proposals by analyzing source files. |
-| `main` | `main()` | - |
-
-## `scripts/sampler.py`
-
-sampler.py — Score and rank sessions by harvest value.
-
-| Function | Signature | Description |
-|----------|-----------|-------------|
-| `scan_session_fast` | `scan_session_fast(path)` | Extract scoring metadata from a session without parsing the full JSONL. |
-| `parse_session_timestamp` | `parse_session_timestamp(filename)` | Parse timestamp from session filename. |
-| `score_session` | `score_session(meta, now, seen_repos)` | Score a session for harvest value. Returns (score, breakdown). |
-| `main` | `main()` | - |
-
-## `scripts/session_metadata.py`
-
-session_metadata.py - Extract structured metadata from Hermes session transcripts.
-
-| Function | Signature | Description |
-|----------|-----------|-------------|
-| `extract_session_metadata` | `extract_session_metadata(file_path)` | Extract structured metadata from a Hermes session JSONL transcript. |
-| `process_session_directory` | `process_session_directory(directory_path, output_file)` | Process all JSONL files in a directory. |
-| `main` | `main()` | CLI entry point. |
-
-## `scripts/session_pair_harvester.py`
-
-Session Transcript → Training Pair Harvester
-
-| Function | Signature | Description |
-|----------|-----------|-------------|
-| `compute_hash` | `compute_hash(text)` | Content hash for deduplication. |
-| `extract_pairs_from_session` | `extract_pairs_from_session(session_data, min_ratio, min_response_words)` | Extract terse→rich pairs from a single session object. |
-| `extract_from_jsonl_file` | `extract_from_jsonl_file(filepath, **kwargs)` | Extract pairs from a session JSONL file. |
-| `deduplicate_pairs` | `deduplicate_pairs(pairs)` | Remove duplicate pairs across files. |
-| `main` | `main()` | - |
-
-## `scripts/session_reader.py`
-
-session_reader.py — Parse Hermes session JSONL transcripts.
-
-| Function | Signature | Description |
-|----------|-----------|-------------|
-| `read_session` | `read_session(path)` | Read a session JSONL file and return all messages as a list. |
-| `read_session_iter` | `read_session_iter(path)` | Iterate over session messages without loading all into memory. |
-| `extract_conversation` | `extract_conversation(messages)` | Extract user/assistant conversation turns, skipping tool-only messages. |
-| `truncate_for_context` | `truncate_for_context(messages, head, tail)` | Truncate long sessions: keep first N + last N messages. |
-| `messages_to_text` | `messages_to_text(messages)` | Convert message list to plain text for LLM consumption. |
-| `get_session_metadata` | `get_session_metadata(path)` | Extract metadata from a session file (first message often has config info). |
-
-## `scripts/test_automation_opportunity_finder.py`
-
-Tests for scripts/automation_opportunity_finder.py — 8 tests.
-
-| Function | Signature | Description |
-|----------|-----------|-------------|
-| `test_analyze_cron_jobs_no_file` | `test_analyze_cron_jobs_no_file()` | Returns empty list when no cron jobs file exists. |
-| `test_analyze_cron_jobs_disabled` | `test_analyze_cron_jobs_disabled()` | Detects disabled cron jobs. |
-| `test_analyze_cron_jobs_errors` | `test_analyze_cron_jobs_errors()` | Detects cron jobs with error status. |
-| `test_analyze_documents_finds_todos` | `test_analyze_documents_finds_todos()` | Detects TODO markers in documents. |
-| `test_analyze_scripts_repeated_commands` | `test_analyze_scripts_repeated_commands()` | Detects repeated shell commands across scripts. |
-| `test_analyze_session_transcripts` | `test_analyze_session_transcripts()` | Detects repeated tool-call sequences. |
-| `test_deduplicate_proposals` | `test_deduplicate_proposals()` | Deduplicates proposals with similar titles. |
-| `test_rank_proposals` | `test_rank_proposals()` | Ranks proposals by impact * confidence. |
-
-## `scripts/test_bootstrapper.py`
-
-Tests for bootstrapper.py — context assembly from knowledge store.
-
-| Function | Signature | Description |
-|----------|-----------|-------------|
-| `make_index` | `make_index(facts, tmp_dir)` | Create a temporary index.json with given facts. |
-| `test_empty_index` | `test_empty_index()` | Empty knowledge store produces graceful output. |
-| `test_filter_by_repo` | `test_filter_by_repo()` | Filter facts by repository. |
-| `test_filter_by_agent` | `test_filter_by_agent()` | Filter facts by agent type. |
-| `test_no_global_flag` | `test_no_global_flag()` | Excluding global facts works. |
-| `test_sort_by_confidence` | `test_sort_by_confidence()` | Facts sort by confidence descending. |
-| `test_sort_pitfalls_first` | `test_sort_pitfalls_first()` | Pitfalls sort before facts at same confidence. |
-| `test_truncate_to_tokens` | `test_truncate_to_tokens()` | Truncation cuts at line boundary. |
-| `test_estimate_tokens` | `test_estimate_tokens()` | Token estimation is reasonable. |
-| `test_build_full_context` | `test_build_full_context()` | Full context with facts renders correctly. |
-| `test_max_tokens_respected` | `test_max_tokens_respected()` | Output respects max_tokens limit. |
-| `test_missing_index_graceful` | `test_missing_index_graceful()` | Missing index.json doesn't crash. |
-
-## `scripts/test_diff_analyzer.py`
-
-Tests for scripts/diff_analyzer.py — 10 tests.
-
-| Function | Signature | Description |
-|----------|-----------|-------------|
-| `test_empty` | `test_empty()` | - |
-| `test_addition` | `test_addition()` | - |
-| `test_deletion` | `test_deletion()` | - |
-| `test_modification` | `test_modification()` | - |
-| `test_rename` | `test_rename()` | - |
-| `test_multiple_files` | `test_multiple_files()` | - |
-| `test_binary` | `test_binary()` | - |
-| `test_to_dict` | `test_to_dict()` | - |
-| `test_context_only` | `test_context_only()` | - |
-| `test_multi_hunk` | `test_multi_hunk()` | - |
-| `run_all` | `run_all()` | - |
-
-## `scripts/test_gitea_issue_parser.py`
-
-Tests for scripts/gitea_issue_parser.py
-
-| Function | Signature | Description |
-|----------|-----------|-------------|
-| `test_basic_parsing` | `test_basic_parsing()` | - |
-| `test_numbered_criteria` | `test_numbered_criteria()` | - |
-| `test_epic_ref_from_body` | `test_epic_ref_from_body()` | - |
-| `test_empty_body` | `test_empty_body()` | - |
-| `test_no_sections` | `test_no_sections()` | - |
-| `test_multiple_sections` | `test_multiple_sections()` | - |
-| `run_all` | `run_all()` | - |
-
-## `scripts/test_harvest_prompt.py`
-
-Test harness for knowledge extraction prompt.
-
-| Function | Signature | Description |
-|----------|-----------|-------------|
-| `validate_knowledge_item` | `validate_knowledge_item(item, idx)` | Validate a single knowledge item. Returns list of errors. |
-| `validate_extraction` | `validate_extraction(data)` | Validate a full extraction result. Returns (is_valid, errors, warnings). |
-| `validate_transcript_coverage` | `validate_transcript_coverage(data, transcript)` | Check that extracted facts are actually supported by the transcript. |
-| `run_tests` | `run_tests()` | Run the built-in test suite. |
-| `validate_file` | `validate_file(filepath)` | Validate an existing extraction JSON file. |
-
-## `scripts/test_harvest_prompt_comprehensive.py`
-
-Comprehensive tests for knowledge extraction prompt.
-
-| Function | Signature | Description |
-|----------|-----------|-------------|
-| `check_prompt_structure` | `check_prompt_structure()` | - |
-| `check_confidence_scoring` | `check_confidence_scoring()` | - |
-| `check_example_quality` | `check_example_quality()` | - |
-| `check_constraint_coverage` | `check_constraint_coverage()` | - |
-| `check_test_sessions` | `check_test_sessions()` | - |
-| `test_prompt_structure` | `test_prompt_structure()` | - |
-| `test_confidence_scoring` | `test_confidence_scoring()` | - |
-| `test_example_quality` | `test_example_quality()` | - |
-| `test_constraint_coverage` | `test_constraint_coverage()` | - |
-| `test_test_sessions` | `test_test_sessions()` | - |
-
-## `scripts/test_harvester_pipeline.py`
-
-Smoke test for harvester pipeline — verifies the full chain:
-
-| Function | Signature | Description |
-|----------|-----------|-------------|
-| `test_session_reader` | `test_session_reader()` | Test that session_reader parses JSONL correctly. |
-| `test_validate_fact` | `test_validate_fact()` | Test fact validation. |
-| `test_deduplicate` | `test_deduplicate()` | Test deduplication. |
-| `test_knowledge_store_roundtrip` | `test_knowledge_store_roundtrip()` | Test loading and writing knowledge index. |
-| `test_full_chain_no_llm` | `test_full_chain_no_llm()` | Test the full pipeline minus the LLM call. |
-
-## `scripts/test_improvement_proposals.py`
-
-Tests for scripts/improvement_proposals.py — 15 tests.
-
-| Function | Signature | Description |
-|----------|-----------|-------------|
-| `test_empty_sessions` | `test_empty_sessions()` | - |
-| `test_no_patterns_on_clean_sessions` | `test_no_patterns_on_clean_sessions()` | - |
-| `test_repeated_error_detection` | `test_repeated_error_detection()` | Same error across 3+ sessions triggers pattern. |
-| `test_repeated_error_threshold` | `test_repeated_error_threshold()` | 2 occurrences should NOT trigger (threshold is 3). |
-| `test_slow_tool_detection` | `test_slow_tool_detection()` | Tool with avg latency > 5000ms across 5+ calls. |
-| `test_fast_tool_not_flagged` | `test_fast_tool_not_flagged()` | Tool under 5000ms avg should not trigger. |
-| `test_failed_retry_detection` | `test_failed_retry_detection()` | 3+ consecutive calls to same tool triggers retry pattern. |
-| `test_manual_process_detection` | `test_manual_process_detection()` | 10+ tool calls with <= 3 unique tools. |
-| `test_generate_proposals_from_patterns` | `test_generate_proposals_from_patterns()` | Proposals generated from waste patterns. |
-| `test_proposal_roi_positive` | `test_proposal_roi_positive()` | ROI weeks should be a positive number for recoverable time. |
-| `test_proposals_sorted_by_impact` | `test_proposals_sorted_by_impact()` | Proposals should be sorted by monthly hours saved (descending). |
-| `test_format_markdown` | `test_format_markdown()` | Markdown output should contain expected sections. |
-| `test_format_json` | `test_format_json()` | JSON output should be valid and parseable. |
-| `test_normalize_error` | `test_normalize_error()` | Error normalization should remove paths and hashes. |
-| `test_cli_integration` | `test_cli_integration()` | End-to-end test: write input JSON, run script, check output. |
-| `run_all` | `run_all()` | - |
-
-## `scripts/test_knowledge_staleness.py`
-
-Tests for scripts/knowledge_staleness_check.py — 8 tests.
-
-| Function | Signature | Description |
-|----------|-----------|-------------|
-| `test_fresh_entry` | `test_fresh_entry()` | - |
-| `test_stale_entry` | `test_stale_entry()` | - |
-| `test_missing_source` | `test_missing_source()` | - |
-| `test_no_hash` | `test_no_hash()` | - |
-| `test_no_source_field` | `test_no_source_field()` | - |
-| `test_fix_hashes` | `test_fix_hashes()` | - |
-| `test_empty_index` | `test_empty_index()` | - |
-| `test_compute_hash_nonexistent` | `test_compute_hash_nonexistent()` | - |
-| `run_all` | `run_all()` | - |
-
-## `scripts/test_priority_rebalancer.py`
-
-Tests for Priority Rebalancer
-
-| Function | Signature | Description |
-|----------|-----------|-------------|
-| `test` | `test(name)` | - |
-| `assert_eq` | `assert_eq(a, b, msg)` | - |
-| `assert_true` | `assert_true(v, msg)` | - |
-| `assert_false` | `assert_false(v, msg)` | - |
-| `make_issue` | `make_issue(**kwargs)` | - |
-
-## `scripts/test_refactoring_opportunity_finder.py`
-
-Tests for scripts/refactoring_opportunity_finder.py — 10 tests.
-
-| Function | Signature | Description |
-|----------|-----------|-------------|
-| `test_complexity_simple_function` | `test_complexity_simple_function()` | Simple function should have low complexity. |
-| `test_complexity_with_conditionals` | `test_complexity_with_conditionals()` | Function with if/else should have higher complexity. |
-| `test_complexity_with_loops` | `test_complexity_with_loops()` | Function with loops should increase complexity. |
-| `test_complexity_with_class` | `test_complexity_with_class()` | Class with methods should count both. |
-| `test_complexity_syntax_error` | `test_complexity_syntax_error()` | File with syntax error should return zeros. |
-| `test_refactoring_score_high_complexity` | `test_refactoring_score_high_complexity()` | High complexity should give high score. |
-| `test_refactoring_score_low_complexity` | `test_refactoring_score_low_complexity()` | Low complexity should give lower score. |
-| `test_refactoring_score_high_churn` | `test_refactoring_score_high_churn()` | High churn should increase score. |
-| `test_refactoring_score_no_coverage` | `test_refactoring_score_no_coverage()` | No coverage data should assume medium risk. |
-| `test_refactoring_score_large_file` | `test_refactoring_score_large_file()` | Large files should score higher. |
-| `run_all` | `run_all()` | - |
-
-## `scripts/test_session_pair_harvester.py`
-
-Tests for session_pair_harvester.
-
-| Function | Signature | Description |
-|----------|-----------|-------------|
-| `test_basic_extraction` | `test_basic_extraction()` | - |
-| `test_filters_short_responses` | `test_filters_short_responses()` | - |
-| `test_skips_tool_results` | `test_skips_tool_results()` | - |
-| `test_deduplication` | `test_deduplication()` | - |
-| `test_ratio_filter` | `test_ratio_filter()` | - |
-
-## `scripts/validate_knowledge.py`
-
-Validate knowledge files and index.json against the schema.
-
-| Function | Signature | Description |
-|----------|-----------|-------------|
-| `validate_fact` | `validate_fact(fact, src)` | - |
-| `main` | `main()` | - |
-
-
----
-
-**Total scripts documented:** 33
-
-*Generated by `scripts/api_doc_generator.py` (Issue #98)*

scripts/api_doc_generator.py

@@ -1,219 +0,0 @@
-#!/usr/bin/env python3
-"""
-API Doc Generator — Issue #98
-
-Scans all Python modules in `scripts/`, extracts their public API surface
-(module docstring + public function signatures + first-line doc summaries),
-and produces a single markdown reference document at `docs/API.md`.
-
-Usage:
-    python3 scripts/api_doc_generator.py          # Write docs/API.md
-    python3 scripts/api_doc_generator.py --check  # Verify docs/API.md is up-to-date
-    python3 scripts/api_doc_generator.py --json   # Emit JSON for downstream tooling
-"""
-
-from __future__ import annotations
-
-import ast
-import os
-import sys
-from datetime import datetime, timezone
-from pathlib import Path
-from typing import TypedDict, List, Optional
-
-
-# ─── Paths ────────────────────────────────────────────────────────────────────
-SCRIPT_DIR = Path(__file__).resolve().parent
-REPO_ROOT = SCRIPT_DIR.parent
-SCRIPTS_DIR = REPO_ROOT / "scripts"
-DOCS_DIR = REPO_ROOT / "docs"
-OUTPUT_PATH = DOCS_DIR / "API.md"
-
-
-# ─── Data structures ───────────────────────────────────────────────────────────
-class FunctionInfo(TypedDict):
-    name: str
-    signature: str
-    summary: str
-
-
-class ModuleInfo(TypedDict):
-    path: str  # relative to repo root, e.g. "scripts/harvester.py"
-    docstring: str
-    functions: List[FunctionInfo]
-
-
-# ─── AST extraction ────────────────────────────────────────────────────────────
-def extract_functions_from_ast(tree: ast.AST, file_rel: str) -> List[FunctionInfo]:
-    """Extract public function names, signatures, and first-line doc summaries."""
-    funcs: list[FunctionInfo] = []
-
-    for node in ast.iter_child_nodes(tree):
-        if not isinstance(node, ast.FunctionDef):
-            continue
-        # Skip private functions
-        if node.name.startswith("_"):
-            continue
-
-        # Build signature: arg1, arg2=default, *args, **kwargs
-        args = []
-        for arg in node.args.args:
-            args.append(arg.arg)
-        if node.args.vararg:
-            args.append(f"*{node.args.vararg.arg}")
-        if node.args.kwarg:
-            args.append(f"**{node.args.kwarg.arg}")
-
-        # Get first line of docstring
-        summary = ""
-        if (node.body and isinstance(node.body[0], ast.Expr) and
-            isinstance(node.body[0].value, ast.Constant) and
-            isinstance(node.body[0].value.value, str)):
-            raw = node.body[0].value.value.strip()
-            summary = raw.split("\n")[0].strip()
-            if len(summary) > 100:
-                summary = summary[:97] + "..."
-
-        funcs.append({
-            "name": node.name,
-            "signature": ", ".join(args),
-            "summary": summary,
-        })
-
-    return funcs
-
-
-def parse_module(filepath: Path) -> Optional[ModuleInfo]:
-    """Parse a Python file and return its module-level docstring and public functions."""
-    try:
-        with open(filepath, "r", encoding="utf-8") as f:
-            source = f.read()
-        tree = ast.parse(source, filename=str(filepath))
-    except Exception as e:
-        print(f"WARNING: Could not parse {filepath}: {e}", file=sys.stderr)
-        return None
-
-    # Module docstring
-    module_doc = ast.get_docstring(tree) or ""
-    module_doc = module_doc.strip().split("\n")[0]  # first line only
-
-    # Public functions
-    functions = extract_functions_from_ast(tree, filepath.name)
-
-    rel = filepath.relative_to(REPO_ROOT)
-    return {
-        "path": str(rel),
-        "docstring": module_doc,
-        "functions": functions,
-    }
-
-
-# ─── Scanning ──────────────────────────────────────────────────────────────────
-def scan_scripts_dir(scripts_dir: Path) -> List[ModuleInfo]:
-    """Scan all .py files in scripts/ and extract API info."""
-    modules: list[ModuleInfo] = []
-    for pyfile in sorted(scripts_dir.glob("*.py")):
-        info = parse_module(pyfile)
-        if info is not None:
-            modules.append(info)
-    return modules
-
-
-# ─── Markdown rendering ─────────────────────────────────────────────────────────
-def render_markdown(modules: List[ModuleInfo]) -> str:
-    """Generate full docs/API.md content from the scanned modules."""
-    lines = [
-        "# Compounding Intelligence — Scripts API Reference",
-        "",
-        f"*Generated: {datetime.now(timezone.utc).strftime('%Y-%m-%d %H:%M UTC')}*",
-        "",
-        "This document auto-documents the public API surface of all scripts",
-        "in `scripts/`. Each section covers one script: module purpose,",
-        "public functions, and their signatures.",
-        "",
-        "---",
-        "",
-    ]
-
-    for mod in modules:
-        rel = mod["path"]
-        name = Path(rel).stem  # e.g. harvester
-        lines.append(f"## `{rel}`")
-        lines.append("")
-        if mod["docstring"]:
-            lines.append(mod["docstring"])
-            lines.append("")
-
-        if mod["functions"]:
-            lines.append("| Function | Signature | Description |")
-            lines.append("|----------|-----------|-------------|")
-            for fn in mod["functions"]:
-                sig = fn["name"] + "(" + fn["signature"] + ")"
-                desc = fn["summary"] or "-"
-                lines.append(f"| `{fn['name']}` | `{sig}` | {desc} |")
-            lines.append("")
-        else:
-            lines.append("*(no public functions — script runs as `main()` only)*")
-            lines.append("")
-
-    lines.extend([
-        "",
-        "---",
-        "",
-        f"**Total scripts documented:** {len(modules)}",
-        "",
-        "*Generated by `scripts/api_doc_generator.py` (Issue #98)*",
-    ])
-    return "\n".join(lines)
-
-
-# ─── JSON output (optional, for automation) ───────────────────────────────────
-def render_json(modules: List[ModuleInfo]) -> str:
-    """Emit machine-readable JSON version of the API reference."""
-    import json
-    payload = {
-        "generated_at": datetime.now(timezone.utc).isoformat(),
-        "generator": "scripts/api_doc_generator.py",
-        "repo": "Timmy_Foundation/compounding-intelligence",
-        "modules": modules,
-    }
-    return json.dumps(payload, indent=2)
-
-
-# ─── Main ──────────────────────────────────────────────────────────────────────
-def main() -> int:
-    import argparse
-    parser = argparse.ArgumentParser(description="Generate API docs for scripts/")
-    parser.add_argument("--check", action="store_true",
-                        help="Exit 1 if docs/API.md is out-of-date")
-    parser.add_argument("--json", action="store_true",
-                        help="Emit JSON to stdout instead of writing markdown")
-    args = parser.parse_args()
-
-    modules = scan_scripts_dir(SCRIPTS_DIR)
-    modules.sort(key=lambda m: m["path"])
-
-    if args.json:
-        print(render_json(modules))
-        return 0
-
-    md = render_markdown(modules)
-
-    if args.check:
-        if OUTPUT_PATH.exists():
-            existing = OUTPUT_PATH.read_text(encoding="utf-8")
-            if existing == md:
-                print("✅ docs/API.md is up-to-date")
-                return 0
-        print("❌ docs/API.md is missing or out-of-date — regenerate with "
-              "`python3 scripts/api_doc_generator.py`", file=sys.stderr)
-        return 1
-
-    DOCS_DIR.mkdir(parents=True, exist_ok=True)
-    OUTPUT_PATH.write_text(md, encoding="utf-8")
-    print(f"✅ Wrote {OUTPUT_PATH}  ({len(modules)} modules documented)")
-    return 0
-
-
-if __name__ == "__main__":
-    sys.exit(main())

scripts/session_pair_harvester.py

@@ -22,114 +22,95 @@ import sys
 from pathlib import Path
 from typing import Optional
 
+from session_reader import extract_conversation, read_session
+
 
 def compute_hash(text: str) -> str:
     """Content hash for deduplication."""
     return hashlib.sha256(text.encode()).hexdigest()[:16]
 
 
-def extract_pairs_from_session(session_data: dict, min_ratio: float = 1.5,
+def extract_pairs_from_conversation(conversation: list, session_id: str, model: str,
+                                min_ratio: float = 1.5,
                                 min_response_words: int = 20) -> list:
-    """Extract terse→rich pairs from a single session object."""
+    """Extract terse→rich pairs from a normalized conversation."""
     pairs = []
-    conversations = session_data.get("conversations", [])
-    session_id = session_data.get("id", "unknown")
-    model = session_data.get("model", "unknown")
-
     seen_hashes = set()
 
-    for i, msg in enumerate(conversations):
-        # Look for assistant/gpt responses
-        if msg.get("from") not in ("gpt", "assistant"):
+    for i, msg in enumerate(conversation):
+        # Look for assistant responses
+        if msg.get('role') != 'assistant':
             continue
 
-        response_text = msg.get("value", "")
+        response_text = msg.get('content', '')
         if not response_text or len(response_text.split()) < min_response_words:
             continue
 
-        # Find the preceding human message
+        # Find the preceding user message
         prompt_text = ""
         for j in range(i - 1, -1, -1):
-            if conversations[j].get("from") == "human":
-                prompt_text = conversations[j].get("value", "")
+            if conversation[j].get('role') == 'user':
+                prompt_text = conversation[j].get('content', '')
                 break
 
         if not prompt_text:
             continue
 
         # Filter: skip tool results, system messages embedded as human
-        if prompt_text.startswith("{") and "output" in prompt_text[:100]:
-            continue  # likely a tool result
-        if prompt_text.startswith("# SOUL.md") or prompt_text.startswith("You are"):
-            continue  # system prompt leak
+        if prompt_text.startswith('{') and 'output' in prompt_text[:100]:
+            continue
+        if prompt_text.startswith('# SOUL.md') or prompt_text.startswith('You are'):
+            continue
 
         # Quality filters
         prompt_words = len(prompt_text.split())
         response_words = len(response_text.split())
 
-        # Must have meaningful length ratio
         if prompt_words == 0 or response_words == 0:
             continue
         ratio = response_words / prompt_words
         if ratio < min_ratio:
             continue
 
-        # Skip responses that are mostly code
-        code_blocks = response_text.count("```")
-        if code_blocks >= 4 and len(response_text.replace("```", "").strip()) < 50:
+        code_blocks = response_text.count('```')
+        if code_blocks >= 4 and len(response_text.replace('```', '').strip()) < 50:
             continue
 
-        # Skip responses with tool call artifacts
-        if "tool_call" in response_text[:100] or "function_call" in response_text[:100]:
+        if 'tool_call' in response_text[:100] or 'function_call' in response_text[:100]:
             continue
 
-        # Deduplicate by content hash
         content_hash = compute_hash(prompt_text + response_text[:200])
         if content_hash in seen_hashes:
             continue
         seen_hashes.add(content_hash)
 
-        # Clean up response: remove markdown headers if too many
         clean_response = response_text
 
         pairs.append({
-            "terse": prompt_text.strip(),
-            "rich": clean_response.strip(),
-            "source": session_id,
-            "model": model,
-            "prompt_words": prompt_words,
-            "response_words": response_words,
-            "ratio": round(ratio, 2),
+            'terse': prompt_text.strip(),
+            'rich': clean_response.strip(),
+            'source': session_id,
+            'model': model,
+            'prompt_words': prompt_words,
+            'response_words': response_words,
+            'ratio': round(ratio, 2),
         })
 
     return pairs
 
 
-def extract_from_jsonl_file(filepath: str, **kwargs) -> list:
-    """Extract pairs from a session JSONL file."""
-    pairs = []
-    path = Path(filepath)
 
-    if not path.exists():
-        print(f"Warning: {filepath} not found", file=sys.stderr)
-        return pairs
-
-    content = path.read_text()
-    lines = content.strip().split("\n")
-
-    for line in lines:
-        line = line.strip()
-        if not line:
-            continue
-        try:
-            session = json.loads(line)
-        except json.JSONDecodeError:
-            continue
-
-        session_pairs = extract_pairs_from_session(session, **kwargs)
-        pairs.extend(session_pairs)
-
-    return pairs
+def extract_from_jsonl_file(path: str, **kwargs) -> list:
+    """Read a session file and extract training pairs using normalized conversation."""
+    session_messages = read_session(path)
+    if not session_messages:
+        return []
+    conversation = extract_conversation(session_messages)
+    # Derive session_id and model from first real message metadata
+    first_msg = next((m for m in session_messages if m.get('role') or m.get('from')), {})
+    session_id = first_msg.get('meta_session_id', Path(path).name)
+    model = first_msg.get('model', 'unknown')
+    return extract_pairs_from_conversation(conversation, session_id, model, **kwargs)
 
 
 def deduplicate_pairs(pairs: list) -> list:

tests/test_api_doc_generator.py

@@ -1,148 +0,0 @@
-#!/usr/bin/env python3
-"""
-Smoke tests for API Doc Generator — Issue #98
-
-Validates that the generator runs, produces docs/API.md, and that
-the generated markdown contains expected sections for the known scripts.
-"""
-
-from __future__ import annotations
-
-import subprocess
-import sys
-from pathlib import Path
-
-import pytest
-
-# Resolve repo root
-REPO_ROOT = Path(__file__).resolve().parents[1]
-SCRIPTS_DIR = REPO_ROOT / "scripts"
-DOCS_DIR = REPO_ROOT / "docs"
-API_MD = DOCS_DIR / "API.md"
-GENERATOR = SCRIPTS_DIR / "api_doc_generator.py"
-
-
-# ─── Generator presence ─────────────────────────────────────────────────────────
-class TestGeneratorPresence:
-    def test_generator_script_exists(self):
-        assert GENERATOR.exists(), f"Missing: {GENERATOR}"
-
-    def test_generator_is_executable(self):
-        with open(GENERATOR) as f:
-            first = f.readline().strip()
-        assert first.startswith("#!"), "Missing shebang"
-        assert "python" in first.lower()
-
-
-# ─── API.md generation ──────────────────────────────────────────────────────────
-class TestAPIDocGeneration:
-    def test_generator_runs_successfully(self):
-        """Run the generator and verify exit code 0."""
-        result = subprocess.run(
-            [sys.executable, str(GENERATOR)],
-            capture_output=True, text=True, cwd=REPO_ROOT, timeout=30
-        )
-        assert result.returncode == 0, (
-            f"Generator failed (code {{result.returncode}})\n"
-            f"STDERR: {{result.stderr[:500]}}"
-        )
-
-    def test_api_md_is_created(self):
-        """docs/API.md must exist after generation."""
-        assert API_MD.exists(), f"Missing output: {API_MD}"
-
-    def test_api_md_is_not_empty(self):
-        """Generate markdown must have substantial content."""
-        content = API_MD.read_text(encoding="utf-8")
-        assert len(content) > 1000, "API.md is suspiciously small"
-
-    def test_api_md_has_expected_structure(self):
-        """Top-level headings and table markers must be present."""
-        content = API_MD.read_text(encoding="utf-8")
-        assert "# Compounding Intelligence — Scripts API Reference" in content
-        assert "## `scripts/" in content
-        assert "| Function | Signature | Description |" in content
-
-    def test_api_md_covers_expected_scripts(self):
-        """At minimum the core scripts should be documented."""
-        content = API_MD.read_text(encoding="utf-8")
-        # Core scripts that must appear
-        core = ["scripts/harvester.py", "scripts/bootstrapper.py",
-                "scripts/session_reader.py", "scripts/dedup.py"]
-        for rel in core:
-            assert f"## `{rel}`" in content, f"Missing section for {rel}"
-
-    def test_api_md_contains_function_names(self):
-        """Spot-check: known public functions from key modules must appear."""
-        content = API_MD.read_text(encoding="utf-8")
-        checks = [
-            ("harvester", "read_session"),
-            ("bootstrapper", "load_index"),
-            ("session_reader", "extract_conversation"),
-            ("dedup", "normalize_text"),
-        ]
-        for module_stem, func_name in checks:
-            assert f"| `{func_name}` |" in content,                 f"Missing function {func_name} from {module_stem}"
-
-
-# ─── Idempotence / --check ─────────────────────────────────────────────────────
-class TestIdempotence:
-    def test_check_flag_passes_when_current(self):
-        """`--check` should exit 0 immediately after generation."""
-        result = subprocess.run(
-            [sys.executable, str(GENERATOR), "--check"],
-            capture_output=True, text=True, cwd=REPO_ROOT, timeout=30
-        )
-        assert result.returncode == 0, (
-            f"--check failed\nSTDOUT: {{result.stdout}}\nSTDERR: {{result.stderr[:200]}}"
-        )
-
-    def test_check_fails_when_api_md_stale(self):
-        """If docs/API.md is manually altered, --check should detect staleness."""
-        # Generate fresh baseline first
-        subprocess.run([sys.executable, str(GENERATOR)], capture_output=True, cwd=REPO_ROOT, timeout=30)
-
-        # Corrupt API.md slightly (append a line at the end)
-        original = API_MD.read_text(encoding="utf-8")
-        corrupted = original + "\n<!-- corrupted -->\n"
-        API_MD.write_text(corrupted, encoding="utf-8")
-
-        # --check should now fail
-        result = subprocess.run(
-            [sys.executable, str(GENERATOR), "--check"],
-            capture_output=True, text=True, cwd=REPO_ROOT, timeout=30
-        )
-        assert result.returncode != 0, "--check should detect stale API.md"
-        assert "out-of-date" in result.stderr.lower() or "out-of-date" in result.stdout.lower()
-
-        # Restore clean state
-        subprocess.run([sys.executable, str(GENERATOR)], capture_output=True, cwd=REPO_ROOT, timeout=30)
-        assert API_MD.read_text(encoding="utf-8") == original
-
-# ─── JSON output ────────────────────────────────────────────────────────────────
-class TestJSONOutput:
-    def test_json_flag_emits_valid_json(self):
-        result = subprocess.run(
-            [sys.executable, str(GENERATOR), "--json"],
-            capture_output=True, text=True, cwd=REPO_ROOT, timeout=30
-        )
-        assert result.returncode == 0
-        import json
-        payload = json.loads(result.stdout)
-        assert "modules" in payload
-        assert len(payload["modules"]) >= 30
-
-    def test_json_has_expected_fields(self):
-        result = subprocess.run(
-            [sys.executable, str(GENERATOR), "--json"],
-            capture_output=True, text=True, cwd=REPO_ROOT, timeout=30
-        )
-        import json
-        payload = json.loads(result.stdout)
-        mod = payload["modules"][0]
-        for key in ("path", "docstring", "functions"):
-            assert key in mod, f"Missing key {{key}} in module payload"
-
-
-if __name__ == "__main__":
-    pytest.main([__file__, "-v"])

tests/test_session_pair_harvester.py

@@ -0,0 +1,118 @@
+"""
+Tests for session_pair_harvester — training pair extraction from sessions.
+"""
+
+import json
+import tempfile
+import unittest
+from pathlib import Path
+
+import sys
+from pathlib import Path
+sys.path.insert(0, str(Path(__file__).parent.parent / "scripts"))
+from session_pair_harvester import (
+    extract_pairs_from_conversation,
+    extract_from_jsonl_file,
+    deduplicate_pairs,
+    compute_hash,
+)
+
+
+class TestSessionPairHarvester(unittest.TestCase):
+    def test_compute_hash_consistent(self):
+        h1 = compute_hash("hello world")
+        h2 = compute_hash("hello world")
+        self.assertEqual(h1, h2)
+        self.assertEqual(len(h1), 16)
+
+    def test_extract_simple_qa_pair(self):
+        """A simple user→assistant exchange produces one pair."""
+        conversation = [
+            {"role": "user", "content": "What is the capital of France?"},
+            {"role": "assistant", "content": "The capital of France is Paris. It is a major European city renowned for its art, fashion, gastronomy, cultural heritage, and historical significance. The city attracts millions of tourists annually."},
+        ]
+        pairs = extract_pairs_from_conversation(conversation, "test_session", "test-model")
+        self.assertEqual(len(pairs), 1)
+        self.assertEqual(pairs[0]["terse"], "What is the capital of France?")
+        self.assertIn("Paris", pairs[0]["rich"])
+        self.assertEqual(pairs[0]["source"], "test_session")
+
+    def test_min_ratio_filter(self):
+        """Very short responses are filtered out."""
+        conversation = [
+            {"role": "user", "content": "Yes"},
+            {"role": "assistant", "content": "No."},
+        ]
+        # Default min_ratio = 1.5, min_words = 20 for response
+        pairs = extract_pairs_from_conversation(conversation, "s", "m", min_response_words=3)
+        self.assertEqual(len(pairs), 0)
+
+    def test_min_words_filter(self):
+        """Assistant responses below min word count are skipped."""
+        conversation = [
+            {"role": "user", "content": "Explain the project architecture in detail"},
+            {"role": "assistant", "content": "OK."},
+        ]
+        pairs = extract_pairs_from_conversation(conversation, "s", "m", min_response_words=5)
+        self.assertEqual(len(pairs), 0)
+
+    def test_skip_non_assistant_messages(self):
+        """System and tool messages are ignored."""
+        conversation = [
+            {"role": "system", "content": "You are a helpful assistant."},
+            {"role": "user", "content": "Hello"},
+            {"role": "assistant", "content": "Hi there! How can I help you today?"},
+        ]
+        pairs = extract_pairs_from_conversation(conversation, "s", "m", min_response_words=3)
+        self.assertEqual(len(pairs), 1)
+        self.assertEqual(pairs[0]["terse"], "Hello")
+
+    def test_multiple_pairs_from_one_session(self):
+        """A conversation with several Q&A turns yields multiple pairs."""
+        conversation = [
+            {"role": "user", "content": "First question?"},
+            {"role": "assistant", "content": "Here is a detailed and comprehensive answer that thoroughly explores multiple aspects of the subject. It provides background context and practical implications for the reader."},
+            {"role": "user", "content": "Second?"},
+            {"role": "assistant", "content": "Another comprehensive response with detailed examples. This includes practical code blocks and thorough explanations to ensure deep understanding of the topic at hand."},
+        ]
+        pairs = extract_pairs_from_conversation(conversation, "s", "m", min_ratio=1.0)
+        self.assertEqual(len(pairs), 2)
+
+    def test_deduplication_removes_duplicates(self):
+        """Identical pairs across sessions are deduplicated."""
+        pairs = [
+            {"terse": "q1", "rich": "a1", "source": "s1", "model": "m"},
+            {"terse": "q1", "rich": "a1", "source": "s2", "model": "m"},
+            {"terse": "q2", "rich": "a2", "source": "s1", "model": "m"},
+        ]
+        unique = deduplicate_pairs(pairs)
+        self.assertEqual(len(unique), 2)
+        sources = {p["source"] for p in unique}
+        # First unique pair can be from either s1 or s2
+        self.assertIn("s1", sources)
+
+    def test_integration_with_test_sessions(self):
+        """Harvester finds pairs in real test session files."""
+        repo_root = Path(__file__).parent.parent
+        test_sessions_dir = repo_root / "test_sessions"
+        if not test_sessions_dir.exists():
+            self.skipTest("test_sessions not found")
+
+        pairs = []
+        for jsonl_file in sorted(test_sessions_dir.glob("*.jsonl")):
+            pairs.extend(extract_from_jsonl_file(str(jsonl_file)))
+
+        self.assertGreater(len(pairs), 0, "Should extract at least one pair from test_sessions")
+        for p in pairs:
+            self.assertIn("terse", p)
+            self.assertIn("rich", p)
+            self.assertIn("source", p)
+            self.assertIn("model", p)
+            # Verify content exists
+            self.assertGreater(len(p["terse"]), 0)
+            self.assertGreater(len(p["rich"]), 0)
+
+
+if __name__ == "__main__":
+    unittest.main()
+