kayba-ai ACE — Learn from Traces
This skill ships learn_from_traces.py, a script that reads OpenClaw session
transcripts, feeds them through the ACE learning pipeline, and writes an
updated skillbook to disk.
Usage
python learn_from_traces.py [OPTIONS] [FILES...]
The script auto-discovers new sessions from ~/.openclaw/agents/<agent>/sessions/
and only processes files that haven't been processed before. Processed filenames
are tracked in ace_processed.txt.
Options
| Flag | Description |
|---|---|
--dry-run |
Parse sessions but skip the learning step (no LLM calls) |
--reprocess |
Ignore the processed log and reprocess all sessions |
--agent ID |
OpenClaw agent ID (default: $OPENCLAW_AGENT_ID or main) |
--output DIR |
Output directory for skillbook files (default: script directory) |
--opik |
Enable Opik observability logging |
Pass one or more JSONL file paths as positional arguments to process specific files instead of auto-discovering sessions.
Examples
Learn from all new sessions
python learn_from_traces.py
Discovers unprocessed sessions under ~/.openclaw/agents/main/sessions/,
runs the learning pipeline, and writes the updated skillbook.
Dry run (no LLM calls)
python learn_from_traces.py --dry-run
Parses and validates sessions without calling the LLM. Useful for checking that session files are readable before committing to a full run.
Process a specific trace file
python learn_from_traces.py ~/.openclaw/agents/main/sessions/f967d602.jsonl
Skips auto-discovery and processes only the given file. The processed log is not updated when files are passed directly.
Process multiple files
python learn_from_traces.py session1.jsonl session2.jsonl session3.jsonl
Reprocess all sessions
python learn_from_traces.py --reprocess
Ignores ace_processed.txt and reprocesses every session file. Useful after
upgrading ACE or when you want to rebuild the skillbook from scratch.
Use a different agent
python learn_from_traces.py --agent my-agent
Looks for sessions in ~/.openclaw/agents/my-agent/sessions/ instead of
the default main.
Write output to a custom directory
python learn_from_traces.py --output /tmp/ace-out
Writes ace_skillbook.json, ace_skillbook.md, and ace_processed.txt
to /tmp/ace-out/ instead of the script's directory.
Enable Opik observability
python learn_from_traces.py --opik
Logs LLM calls and pipeline steps to Opik for debugging and monitoring.
Run from the ACE repo (host setup)
cd /path/to/agentic-context-engine
uv run python examples/openclaw/kayba-ace/learn_from_traces.py --dry-run
When running from a repo checkout, uv run ensures ace is importable.
Run inside Docker
# Dry run
docker run --rm -v ~/.openclaw:/home/node/.openclaw openclaw:local ace-learn --dry-run
# Full run with API key
docker run --rm \
-v ~/.openclaw:/home/node/.openclaw \
-e AWS_BEARER_TOKEN_BEDROCK="$AWS_BEARER_TOKEN_BEDROCK" \
openclaw:local ace-learn
Inside the extended Docker image, ace-learn is a wrapper that calls this
script with the correct paths.
Environment Variables
| Variable | Default | Description |
|---|---|---|
ACE_MODEL |
bedrock/us.anthropic.claude-sonnet-4-20250514-v1:0 |
LLM model for reflection and skill extraction |
OPENCLAW_AGENT_ID |
main |
Default agent ID (overridden by --agent) |
OPENCLAW_HOME |
~/.openclaw |
OpenClaw home directory |
The script also loads .env files from $OPENCLAW_HOME/.env and ~/.env.
Set your API key in one of these variables:
AWS_BEARER_TOKEN_BEDROCKANTHROPIC_API_KEYOPENROUTER_API_KEYLITELLM_API_KEYSPH_LITELLM_KEY
Output Files
| File | Description |
|---|---|
ace_skillbook.json |
Machine-readable skillbook (persists across runs) |
ace_skillbook.md |
Human-readable skillbook loaded by the agent |
ace_processed.txt |
Tracks which sessions have been processed |
Setup
Run the setup script from the ACE repo to copy this skill into your OpenClaw workspace:
python examples/openclaw/setup.py
Full guide: https://kayba-ai.github.io/agentic-context-engine/integrations/openclaw/