Session Scribe

Automatically summarize active OpenClaw session transcripts into daily memory files using a cheap LLM. Run as a system cron job — reads new transcript entrie...

Security Scan

The skill's code and instructions match its stated purpose (scribing session transcripts to daily memory files via an external LLM), with no obvious misdirection or hidden endpoints — but it will read all OpenClaw session transcripts you point it at and send their contents to an external API, so treat API keys and transcript contents as sensitive.

Purpose & Capability

Name/description (summarize session transcripts into daily memory files) align with what the code does: reading ~/.openclaw sessions, extracting user/assistant turns, calling OpenAI/Anthropic APIs, and appending summaries to a memory directory. Required binary (node) is appropriate.

Instruction Scope

Instructions and the script operate on OpenClaw session directories and sessions.json as described. This is consistent with purpose but has broad read scope: in --all-sessions mode the script iterates agents and will consider many transcripts. The SKILL.md includes a clear privacy notice; you should ensure transcripts do not contain secrets before enabling cron runs.

Install Mechanism

This is instruction-only (no install spec). SKILL.md tells users to run npm install in the skill directory, but the included scribe.js uses only Node core modules (fs/path/https) and there is no package.json in the manifest — npm install appears unnecessary and may be misleading. No remote downloads or obscure install sources are present.

Credentials

Registry lists OPENAI_API_KEY as the required env var and primary credential, which is coherent but incomplete: the script also accepts ANTHROPIC_API_KEY and an --api-key-file. The script only needs an LLM API key (or key file); it does not request unrelated credentials. Ensure you use a low-privilege key and limit file permissions.

Persistence & Privilege

always is false and the skill does not modify other skills or system-wide settings. It writes its own .scribe-state.json (kept in the skill directory) and appends memory files to the configured memory directory; this is normal for its function.

Assessment

This skill appears to do exactly what it claims: read OpenClaw transcripts, call an external LLM, and append bullet-point summaries to daily memory files. Important things to consider before installing: - Privacy/exfiltration risk: transcripts (including any secrets, tokens, PII) will be sent to the configured LLM provider. Only enable cron runs if you are comfortable with that. Use a separate low‑privilege API key with spend limits. - Test first: run the provided --dry-run option to see exactly what content would be sent and what would be written before enabling cron. - API key handling: the script supports OPENAI_API_KEY, ANTHROPIC_API_KEY, or an --api-key-file. The registry metadata only lists OPENAI_API_KEY — if you plan to use Anthropic or a file, set the appropriate environment or file. Store key files with restrictive permissions (chmod 600) as suggested. - npm install note: SKILL.md recommends npm install, but the included scribe.js appears to use only Node core modules and no package.json is present. Inspect the skill directory before running npm install to avoid pulling unexpected packages. - Scope review: when using --all-sessions or pointing agents dir at ~/.openclaw/agents, the script may read many transcripts. If you only want a single session, use --session-id or --auto-session to limit exposure. If these tradeoffs are acceptable (you accept sending transcripts to an external provider and take key-management precautions), the skill is coherent with its stated purpose.

Reads OpenClaw session transcripts and writes summarized bullet points to a daily memory file — automatically, via system cron.

No gateway involvement. No context bloat. Just cheap, reliable scribing.

⚠️ Privacy notice

This skill sends conversation transcript content to an external LLM (OpenAI or Anthropic). If your sessions contain secrets, API keys, PII, or sensitive data — be aware that content will be sent to the provider for summarization. Use a dedicated low-privilege API key with spend limits, and consider running --dry-run first to see exactly what gets sent.

How it works

1. Reads <sessions-dir>/<session-id>.jsonl for new entries since last run
2. Extracts user + assistant turns (skips tool calls, system events)
3. Sends to OpenAI or Anthropic API for summarization (gpt-4o-mini by default — cheap)
4. Appends bullet-point summary to <memory-dir>/YYYY-MM-DD.md (date auto-generated)
5. Saves progress in .scribe-state.json so next run only processes new entries

Setup

# Install dependencies (one-time, in the skill directory)
cd /path/to/skills/session-scribe && npm install

# Set your API key — OpenAI or Anthropic both work
export OPENAI_API_KEY=sk-...
# or
export ANTHROPIC_API_KEY=sk-ant-api03-...

# Find your session ID
python3 -c "
import json, os
sessions = json.load(open(os.path.expanduser('~/.openclaw/agents/main/sessions/sessions.json')))
for k, v in sessions.items():
    print(k, '->', v.get('sessionId'))
"

# Test run (dry-run, prints summary without writing)
node scripts/scribe.js \
  --sessions ~/.openclaw/agents/main/sessions \
  --session-id <your-session-id> \
  --memory-dir ~/.openclaw/workspace/memory \
  --dry-run

# Or auto-resolve session by key suffix (e.g. your Discord channel ID)
node scripts/scribe.js \
  --sessions ~/.openclaw/agents/main/sessions \
  --auto-session "discord:channel:YOUR_CHANNEL_ID" \
  --memory-dir ~/.openclaw/workspace/memory \
  --dry-run

Note: OpenClaw's internal oat tokens cannot be used for direct API calls. Use a standard API key from platform.openai.com or console.anthropic.com.

System cron setup

Add to crontab (crontab -e) to run every hour:

# Store your key in a protected file (do NOT inline secrets in crontab)
echo "sk-your-key" > ~/.openclaw/secrets/scribe-key
chmod 600 ~/.openclaw/secrets/scribe-key

# Single session (via key file)
0 * * * * node /path/to/session-scribe/scripts/scribe.js \
  --sessions ~/.openclaw/agents/main/sessions \
  --auto-session "discord:channel:YOUR_CHANNEL_ID" \
  --memory-dir ~/.openclaw/workspace/memory \
  --api-key-file ~/.openclaw/secrets/scribe-key \
  >> /tmp/scribe.log 2>&1

# All sessions across all agents
0 * * * * node /path/to/session-scribe/scripts/scribe.js \
  --agents-dir ~/.openclaw/agents \
  --all-sessions \
  --memory-dir ~/.openclaw/workspace/memory \
  --api-key-file ~/.openclaw/secrets/scribe-key \
  >> /tmp/scribe.log 2>&1

Options

Flag	Description	Default
--sessions <dir>	Path to OpenClaw sessions directory	required
--session-id <id>	Specific session UUID to scribe	—
--auto-session <key>	Auto-resolve session ID from sessions.json key suffix	—
--memory-dir <dir>	Directory to write daily memory files	required
--provider <name>	LLM provider: openai or anthropic	auto-detected from env
--model <model>	Model to use	gpt-4o-mini
--api-key <key>	API key (alternative to env var)	—
--api-key-file <path>	Read API key from file	—
--agent <id>	Agent label shown in memory file headers	agent
--dry-run	Print summary without writing to disk	false
--min-turns <n>	Minimum new turns before scribing	3
--active-within-hours <n>	Only scribe sessions active within this window (--all-sessions mode)	1

Pairing with supermemory

For a full automated memory pipeline:

1. session-scribe runs hourly → appends to memory/YYYY-MM-DD.md
2. supermemory ingest runs every 2h → syncs changed files to Supermemory
3. supermemory recall at session start → enriched context from past sessions

References

• references/transcript-format.md — OpenClaw JSONL transcript structure and sessions.json format