Purpose
Run the target-repo-skills experiment to determine whether a target repository's .claude/skills/ are discovered by the fullsend triage agent inside an OpenShell sandbox.
Requirements
| Requirement | Link |
|---|---|
| fullsend CLI | Built from fullsend repo: go build ./cmd/fullsend |
| gh CLI | https://cli.github.com/ |
| git | https://git-scm.com/ |
| GCP credentials | gcloud auth application-default login |
Environment variables
| Variable | Description |
|---|---|
FULLSEND_DIR | Path to the fullsend scaffold directory (e.g., /path/to/fullsend/internal/scaffold/fullsend-repo) |
ANTHROPIC_VERTEX_PROJECT_ID | GCP project ID with Vertex AI access |
CLOUD_ML_REGION | GCP region for Vertex AI (e.g., us-east5) |
GOOGLE_APPLICATION_CREDENTIALS | Path to GCP service account key or ADC file |
TARGET_ORG | (Optional) GitHub org/user for the synthetic repo. Defaults to maruiz93 |
TARGET_REPO | (Optional) Repo name. Defaults to experiment-target-repo-skills |
Steps
Navigate to the experiment directory:
bashcd target-repo-skillsSet required environment variables:
bashexport FULLSEND_DIR=/path/to/fullsend/internal/scaffold/fullsend-repo export ANTHROPIC_VERTEX_PROJECT_ID=your-project-id export CLOUD_ML_REGION=us-east5 export GOOGLE_APPLICATION_CREDENTIALS=~/.config/gcloud/application_default_credentials.jsonCreate the synthetic target repo and file the test issue:
bash./setup-target-repo.shRun the experiment:
bash./run.sh
Expected Output
results/control/agent-result.json— triage output without the skillresults/control/transcript.jsonl— Claude session transcript without the skillresults/treatment/agent-result.json— triage output with the skillresults/treatment/transcript.jsonl— Claude session transcript with the skill- The treatment transcript should contain
triage-guidancein the system prompt if skill discovery works - The treatment
agent-result.jsonshould use labels from the skill's taxonomy (area:api,priority:critical,type:bug)
Viewing Transcripts
Use claude-replay to view the session transcripts as interactive HTML replays:
bash
# Control run (no skill)
claude-replay results/control/transcript.jsonl --title "Control Run" --serve
# Treatment run (with triage-guidance skill)
claude-replay results/treatment/transcript.jsonl --title "Treatment Run" --serveOr generate static HTML files:
bash
claude-replay results/control/transcript.jsonl --title "Control Run" -o control-replay.html
claude-replay results/treatment/transcript.jsonl --title "Treatment Run" -o treatment-replay.html