auto-capture-claude-code
/auto-capture-claude-code
Claude Code adapter for the auto-capture skill. Extends auto-capture with automatic session-end hooks that capture transcripts to Open Brain without manual intervention. Use this when you want every meaningful Claude Code session to be preserved automatically — not just the ones where you remember to say "wrap up".
- Author
- natebjones-projects ↗
- Repository
- ob1 ↗
- Licence
- MIT
- Last repository change
- 10.09.2026
Assessment and security
- Files in this package
- 4
- Scripts
- 1
- HTML comments
- 0
- URLs in SKILL.md
- 1
No signals detected. Repository changed on 10.09.2026, without references to retired models or APIs.
MIT, open: use and modify while retaining author attribution.
- 1 scripts in this package. Review them before installing, as Claude may execute them.
Scripts: session-end-capture.mjs
What it does and when to use it
Claude Code adapter for the auto-capture skill. Extends auto-capture with automatic session-end hooks that capture transcripts to Open Brain without manual intervention. Use this when you want every meaningful Claude Code session to be preserved automatically — not just the ones where you remember to say "wrap up".
Triggers from the description
Original author description
Claude Code adapter for the auto-capture skill. Extends auto-capture with automatic session-end hooks that capture transcripts to Open Brain without manual intervention. Use this when you want every meaningful Claude Code session to be preserved automatically — not just the ones where you remember to say "wrap up".
How it works
- Relationship to Upstream Skill
- Problem
- What This Adapter Adds
- How It Works
- Skip Heuristics
- Installation
- Adapting the Script
- Output
- Notes
SKILL.md preview
---
name: auto-capture-claude-code
description: |
Claude Code adapter for the auto-capture skill. Extends auto-capture with
automatic session-end hooks that capture transcripts to Open Brain without
manual intervention. Use this when you want every meaningful Claude Code
session to be preserved automatically — not just the ones where you
remember to say "wrap up".
author: Alan Shurafa
version: 1.0.0
requires_skills:
- auto-capture
---
# Auto-Capture: Claude Code Adapter
## Relationship to Upstream Skill
This adapter implements the session-end capture behavior defined by the upstream [auto-capture skill](../auto-capture/) by **Jared Irish**. The base skill is a behavioral protocol — it describes when and what to capture during interactive session closes. This adapter is the concrete Claude Code binding: a Stop-hook script that fires the same capture behavior automatically when a session ends without a verbal trigger. The upstream skill and this adapter are complementary; install both for full coverage.
## Problem
The base auto-capture skill requires a verbal trigger ("wrap up", "park this")
to fire. In practice, many Claude Code sessions end without that trigger —
the user closes the terminal, hits Ctrl+C, or simply walks away. Those sessions
and their decisions are lost.
## What This Adapter Adds
This adapter uses Claude Code's hook system to run a capture script automatically
at session end. It complements the base auto-capture skill:
- **Base auto-capture** handles interactive session-close captures (ACT NOW items,
session summaries) when the user explicitly wraps up.
- **This adapter** handles ambient capture when the session ends without an
explicit trigger (terminal close, timeout, Ctrl+C).
Together they ensure no valuable session falls through the cracks.
## How It Works
1. Claude Code fires a `Stop` hook at session end, passing the transcript path
and session metadata via stdin as JSON.
2. The hook script (`session-end-capture.mjs`) reads the transcript, filters out
short or agent-only sessions, and formats the content.
3. The formatted transcript is POSTed to the Open Brain REST ingest endpoint
(or smart-ingest edge function) for automatic thought extraction.
4. Failed captures are saved to a local retry queue and retried on subsequent
session ends.
## Skip Heuristics
Not every session is worth capturing. The hook skips:
- Sessions with fewer than 3 user turns (too short to contain decisions)
- Agent-only sessions (sub-agent work, automated tooling)
- Sessions containing restricted content (matched against sensitivity patterns)
- Session-end reasons that are not terminal (`clear`, `resume`)
## Installation
### Prerequisites
Install the base [auto-capture skill](../auto-capture/) first. This adapter
extends it — it does not replace it.
### Steps
1. Copy `session-end-capture.mjs` to your project or a shared scripts directory.
2. Register the hook in your Claude Code settings (`.claude/settings.json` or
global `~/.claude/settings.json`):
```json
{
"hooks": {
"Stop": [
{
"matcher": "",
"hooks": [
{
"type": "command",
"command": "node /path/to/session-end-capture.mjs"
}
]
}
]
}
}
```
3. Set environment variables (in `.env.local` or your environment):
```bash
SUPABASE_URL=https://<project-ref>.supabase.co
MCP_ACCESS_KEY=your-access-key
```
4. Restart Claude Code to pick up the hook.
5. End a test session with at least 3 user messages and verify a capture appears
in your Open Brain thoughts.
## Adapting the Script
The included `session-end-capture.mjs` is a reference implementation. Adapt it
to your setup:
- **Ingest endpoint**: Update the URL construction if your REST API or
smart-ingest function is deployed at a different path.
- **Sensitivity patterns**: Add a `config/sensitivity-patterns.json` file with
regex patterns for restricted content detection, or remove the check if you
don't need it.
- **Retry queue**: The script saves failed captures to
`data/capture-retry-queue/` and retries them on subsequent runs. Adjust
`RETRY_MAX_ATTEMPTS` and `RETRY_BATCH_SIZE` as needed.
- **Hard timeout**: The script exits after 25 seconds to avoid blocking Claude
Code shutdown. Adjust `HARD_TIMEOUT_MS` if your network is slower.
## Output
When working correctly:
- Every meaningful Claude Code session (3+ user turns, non-agent, non-restricted)
is automatically ingested into Open Brain for thought extraction.
- Failed captures are queued locally and retried on subsequent session ends.
- Short, agent, and restricted sessions are silently skipped.
- All outcomes are logged to `logs/ambient-capture.log` for debugging.
## Notes
- This adapter is designed to be non-blocking. All errors are caught and logged —
the hook never prevents Claude Code from shutting down.
- The base auto-capture skill and this adapter are complementary. The skill
handles interactive captures with ACT NOW items; the adapter handles ambient
background capture of the full session transcript.
- Tool names vary by client and connector. The hook script uses the REST API
directly rather than MCP tools, so it works regardless of which MCP connector
is active.
Files in this package (4)
- README.md 6 KB
- SKILL.md 5 KB
- metadata.json 1 KB
- session-end-capture.mjs 14 KB
Versions
The database has no other skills named auto-capture-claude-code.
Database snapshot: 13.11.2026 · package version f758aa316fa0 (SHA-256 digest of the entire package)
Downloads in the last 30 days: 0 · ranking
Report a problem with this skill: harmful content, incorrect descriptions or an author’s removal request.
Similar skills
- 90auto-capture
Automatically capture ACT NOW items and a session summary to Open Brain when a work session is ending. Use when wrapping a brainstorm, parking a project, finishing a Panning for Gold run, or otherwise closing a session with decisions worth remembering. Use the Open Brain capture… - 100brain
Capture a thought into the user's personal brain (an Obsidian vault) or look something up there, from any Claude Code session. Only when the user types /brain. - 100auto-skill
Evaluate current session for skill-worthy workflows and create reusable skills. Triggers on: auto-skill, create skill from session, save workflow, capture this as a skill. - 90brain:export
Use when the user asks to refresh CLAUDE.local.md, regenerate project context, or after a brain-update session flush. Rewrites only the sentinel-delimited auto-generated section from the brain DB, preserving manual behavior rules. - 100claude-code-hook-development
This skill should be used when the user asks to "create a hook", "add a hook", "write a hook", or mentions Claude Code hooks. Also suggest this skill when the user asks to "automatically do X" or "run X before/after Y" as these are good candidates for hooks. - 100brain-recall
Persistent recall over the user's saved Claude and Claude Code transcripts in ~/claude-brain, via an always-on vector (RAG) search server that answers in milliseconds. Use when the user asks "what did we decide about…", "how did we do…", "last time we…", "do you remember…",…