afterpaths 0.1.0__tar.gz

afterpaths-0.1.0/.afterpaths/.gitignore

@@ -0,0 +1 @@
+*

afterpaths-0.1.0/.afterpaths/meta.json

@@ -0,0 +1,29 @@
+{
+  "sessions": [
+    {
+      "session_id": "7faf6980-c5cf-41b4-959e-2f889615c9e2",
+      "source": "claude_code",
+      "source_path": "/Users/burnssa/.claude/projects/-Users-burnssa-Code-afterpaths/7faf6980-c5cf-41b4-959e-2f889615c9e2.jsonl",
+      "summary_path": "/Users/burnssa/Code/afterpaths/.afterpaths/summaries/7faf6980-c5cf-41b4-959e-2f889615c9e2.md",
+      "git_refs": [],
+      "created_at": "2026-01-10T15:32:29.232741",
+      "updated_at": "2026-01-10T16:01:21.047440"
+    },
+    {
+      "session_id": "78a3e6cb-9e11-47c5-b4a7-6abc84ffe802",
+      "source": "claude_code",
+      "source_path": "/Users/burnssa/.claude/projects/-Users-burnssa-Code-afterpaths/78a3e6cb-9e11-47c5-b4a7-6abc84ffe802.jsonl",
+      "summary_path": "/Users/burnssa/Code/afterpaths/.afterpaths/summaries/78a3e6cb-9e11-47c5-b4a7-6abc84ffe802.md",
+      "git_refs": [],
+      "created_at": "2026-01-14T20:16:08.674304"
+    }
+  ],
+  "version": 1,
+  "distill": {
+    "sessions_included": [
+      "7faf6980-c5cf-41b4-959e-2f889615c9e2"
+    ],
+    "version": 1,
+    "last_run": "2026-01-10T16:21:53.965064"
+  }
+}

afterpaths-0.1.0/.afterpaths/summaries/78a3e6cb-9e11-47c5-b4a7-6abc84ffe802.md

@@ -0,0 +1,34 @@
+# Removing Paywall and Adding Cursor Support to Afterpaths
+
+**Date:** 2026-01-14 | **Duration:** 6h 38m
+**Files:** README.md, cli.py, licensing.py, base.py, claude_code.py, cursor.py
+
+## Summary
+Removed the Pro licensing paywall from the rules extraction feature, making it free for all users. Added a new Cursor IDE session adapter to read chat history from SQLite databases. Fixed several bugs in session filtering and improved the CLI output to show summarization status.
+
+## Discoveries
+- **Claude Code auto-deletes sessions after 30 days by default** - Users can prevent this by setting `"cleanupPeriodDays": 99999` in `~/.claude/settings.json`. This is a key value prop for afterpaths.
+- **Claude Code automatically loads all `.md` files from `.claude/rules/`** - No code needed to inject rules; just write files to that directory and they're included in context at session start, same priority as `CLAUDE.md`.
+- **Cursor stores chat history in SQLite databases** - Located at `~/Library/Application Support/Cursor/User/workspaceStorage/*/state.vscdb`. The chat data is in tables like `ItemTable` with keys containing "aiChat" or "composer" (Cursor 0.43+).
+- **The `get_sessions_for_cwd()` function only queried Claude Code** - It was hardcoded to use `ClaudeCodeAdapter` instead of iterating through all registered adapters, so Cursor sessions were invisible without `--all`.
+
+## Dead Ends
+- **Substring matching for project filtering** - The original filter checked `if project_filter in project_name`, which caused `/Users/burnssa/Code/superjective` to match both `superjective` and `superjective-extension` directories. Changed to exact path matching with `==`.
+
+## Decisions
+- **Made rules extraction free instead of Pro-only** - User wants to understand usage patterns before deciding what to monetize. Kept the licensing infrastructure for a future "Vault" feature.
+- **Used exact path matching for project filtering** - Substring matching caused cross-project pollution. Exact match is the expected behavior when scoping to a directory.
+- **Removed `activate` command entirely** - No current paid features, so the command was confusing. The `licensing.py` file remains for future use.
+
+## Gotchas & Warnings
+- **Session type filtering hides most sessions by default** - `afterpaths log` only shows "main" sessions, not "agent" sessions (spawned by Task tool). A project might have 17 total sessions but only 4 main ones. Use `--type=all` to see everything.
+- **Summary titles come from two sources** - Afterpaths summaries have a `# heading` that gets extracted. Claude Code also stores its own auto-generated summary in the JSONL. The CLI now shows afterpaths summary if available, otherwise falls back to Claude's.
+- **Cursor workspace folders are stored as JSON in SQLite** - The `folder` key in `state.vscdb` contains a JSON string with a `path` field, not a direct path string.
+
+## Open Questions
+- **What triggers Claude Code's session cleanup?** - Is it strictly 30 days from last modification, or are there other criteria? Some users report unexpected deletions.
+- **How should Cursor Composer chats be differentiated from regular chats?** - The adapter finds both but they may have different structures in Cursor 0.43+.
+- **Should afterpaths proactively warn users about the 30-day retention?** - Could add a check on first run to suggest the settings.json change.
+
+---
+*Generated by anthropic/claude-opus-4-5-20251101 | Tokens: 11620 in, 862 out*

afterpaths-0.1.0/.afterpaths/summaries/7faf6980-c5cf-41b4-959e-2f889615c9e2.md

@@ -0,0 +1,52 @@
+# Building Afterpaths: AI Session Analysis Tool with Git Integration
+
+**Date:** 2026-01-10 | **Duration:** 2h 4m
+**Files:** 13 files modified
+**Git refs:** None detected
+
+## Summary
+Built "Afterpaths" - a tool to extract, analyze and summarize AI coding sessions from Claude Code. Started with session listing/viewing, added git ref extraction and commit tracing via diff matching, then implemented LLM-powered summarization with provider-agnostic architecture. Extended with file modification tracking, repo-scoped git ref filtering, and protective flags for summary management.
+
+## Discoveries
+- **Claude Code session format differs from spec**: Directory names use hyphen-separated paths (`-Users-burnssa-Code-afterpaths`), not base64 encoding as documented
+- **Tool results are embedded**: Tool responses appear as `tool_result` content blocks within `user` entries, not as separate entry types
+- **Agent vs main sessions pattern**: Sessions starting with `agent-` are sub-processes spawned by Task tool, typically 1-2KB vs main sessions of 100KB+
+- **Time filtering complexity**: Session file modification time != session activity time - needed 1-day buffer after commit to catch active sessions
+- **Path decoding ambiguity**: Hyphens in directory names make decoding tricky - added fallback verification against actual filesystem paths
+- **Caching provides 4x speedup**: File modification extraction went from 0.277s to 0.065s with JSON caching
+- **Git ref false positives from tool output**: Command outputs like `afterpaths link` results were being parsed as legitimate git refs - needed to filter against actual repo existence
+- **Sessions can have built-in summaries**: Claude Code itself generates `summary` fields for some sessions, stored directly in the JSONL file
+
+## Dead Ends
+- **Base64 path decoding approach** - Original spec assumed base64 encoding, but actual format uses simple hyphen replacement (discovered after implementing decoder)
+- **30-day commit search window** - Too broad, most commits happen within days of session, reduced to 7 days default
+- **Exact content matching for commits** - Initially planned fuzzy diff matching, but file-path-based matching proved sufficient and much simpler
+- **Git ref extraction from all text** - Initially parsed all session content for refs, but this captured false positives from discussions about other repos and from afterpaths' own output
+
+## Decisions
+- **Session type filtering by default** - Show only main sessions in `log` command since agent sessions are usually noise, but allow `--type=all` for full view
+- **Provider-agnostic LLM architecture** - Built abstraction layer supporting Anthropic, OpenAI, and OpenAI-compatible APIs rather than hardcoding Anthropic
+- **Caching at file-activity level** - Cache extracted file modifications per session rather than raw parsing, provides good speedup without excessive storage
+- **7-day commit trace window** - Reasonable default covering weekend delays while avoiding false positives from old sessions
+- **Repo-scoped git ref filtering** - Filter detected refs against current git repo using `git branch -a` and `git rev-parse --verify` to eliminate false positives from other projects
+- **Summary overwrite protection** - Require `--force` to overwrite existing summaries, `--update` to refine them with LLM
+- **Extract refs from git commands only** - Changed from parsing all text to only extracting from actual Bash tool inputs containing git commands, reducing false positives
+
+## Gotchas & Warnings
+- **Session numbering consistency** - `show` command must use same type filtering as `log` so session numbers match between commands
+- **Git ref false positives** - Documentation text can contain words like "branch" and "commit" - needed context-aware regex filtering
+- **Time zone handling** - Commit times and session times may be in different zones, using 1-day buffer to handle edge cases
+- **Large session files** - Some sessions are 30MB+ (especially hex-line-assignment), parsing can be slow without caching
+- **Cross-repo git ref pollution** - Sessions discussing multiple repos can show refs from other projects unless filtered by current repo
+- **Model version defaults** - Initial config used `claude-sonnet-4-20250514` instead of correct `claude-sonnet-4-5-20250929`; Opus 4.5 is `claude-opus-4-5-20251101`
+- **Renamed directories** - Sessions may reference paths that no longer exist (e.g., `hex-line-assignment` renamed to `hex_line_assignment`) - fallback decoding handles this gracefully
+
+## Open Questions
+- **Content-based commit matching** - Current approach matches by files modified, but doesn't verify actual content similarity - would fuzzy diff matching catch more cases?
+- **Session activity time extraction** - Could parse session timestamps more precisely rather than using file modification time?
+- **Cross-session linking** - Sessions sometimes reference other sessions or continue work - worth extracting these connections?
+- **Summary quality evaluation** - How to measure if generated summaries are actually useful for future work?
+- **Git ref scope balance** - Current repo-only filtering prevents false positives but might miss legitimate cross-repo references
+
+---
+*Generated by anthropic/claude-opus-4-5-20251101 | Tokens: 27721 in, 1196 out*

afterpaths-0.1.0/.env.example

@@ -0,0 +1,33 @@
+# Afterpaths Configuration
+# Copy this file to .env and fill in your values
+
+# =============================================================================
+# LLM Provider Configuration
+# =============================================================================
+
+# Provider: anthropic, openai, or openai-compatible
+AFTERPATHS_LLM_PROVIDER=anthropic
+
+# API Key for your chosen provider
+ANTHROPIC_API_KEY=sk-ant-...
+# OPENAI_API_KEY=sk-...
+
+# Model to use for summarization
+# Anthropic: claude-sonnet-4-5-20250929, claude-opus-4-5-20251101, claude-haiku-3-5-20241022
+# OpenAI: gpt-4o, gpt-4o-mini, gpt-4-turbo
+AFTERPATHS_MODEL=claude-sonnet-4-5-20250929
+
+# =============================================================================
+# OpenAI-Compatible API (for local models, Azure, etc.)
+# =============================================================================
+
+# Base URL for OpenAI-compatible APIs (e.g., local Ollama, LM Studio, Azure)
+# OPENAI_API_BASE=http://localhost:11434/v1
+# OPENAI_API_KEY=not-needed-for-local
+
+# =============================================================================
+# Optional Settings
+# =============================================================================
+
+# Max tokens for summary generation (default: 3000)
+# AFTERPATHS_MAX_TOKENS=3000

afterpaths-0.1.0/.gitignore

@@ -0,0 +1,72 @@
+# Environment and secrets
+.env
+.env.local
+.env.*.local
+*.pem
+*.key
+secrets/
+
+# macOS
+.DS_Store
+.AppleDouble
+.LSOverride
+._*
+
+# Claude Code
+.claude/
+
+# Python
+__pycache__/
+*.py[cod]
+*$py.class
+*.so
+.Python
+build/
+develop-eggs/
+dist/
+downloads/
+eggs/
+.eggs/
+lib/
+lib64/
+parts/
+sdist/
+var/
+wheels/
+*.egg-info/
+.installed.cfg
+*.egg
+
+# Virtual environments
+venv/
+ENV/
+env/
+.venv/
+
+# IDE and editors
+.idea/
+.vscode/
+*.swp
+*.swo
+*~
+.project
+.pydevproject
+.settings/
+
+# Testing and coverage
+.tox/
+.nox/
+.coverage
+.coverage.*
+htmlcov/
+.pytest_cache/
+.hypothesis/
+
+# Logs
+*.log
+logs/
+
+# Misc
+.cache/
+*.bak
+*.tmp