Your first measured coding task.
Download the self-contained early-access plugin. It includes both marketplace manifests and runs locally with Node.js 24+ and Git; no access to Refario’s source repository is needed.
Verified.
Accepted.
Both checks matter.
- Objective verification
- Passed
- Human acceptance
- Yes
- Observed failed tools
- 2
- Monetary cost
- Unavailable
Clarify the failed tool’s usage instructions, then compare the next similar task.
Sample evidence, not a customer result or measured improvement.
Choose your agent host.
Extract the ZIP and open a terminal in the extracted refario-agent-performance directory. Run the commands for your host, then start a fresh agent session.
Codex
codex plugin marketplace add .
codex plugin add refario-agent-performance@refarioOpen /hooks in Codex to review and trust the Refario hooks, then start a fresh session. Installing the plugin alone does not enable its native capture hooks.
Claude Code
claude plugin marketplace add .
claude plugin install refario-agent-performance@refarioSign in to Claude Code before running your first measured task.
Use a current host version with plugin support. Download SHA-256 checksum.
Ask for a measured task.
Use Refario to measure this task and recommend one evidence-backed improvement. Name the Git repository, the change you want, and what a correct result looks like.
Keep the capture ID with your task
Version 0.1.2 binds each review to an exact capture ID. Native hooks supply a session ID for start; retain the capture ID printed by start and use it for review with the same task and repository. Mismatched or ambiguous captures stop with an error. Set REFARIO_DATA_DIR inside your writable project before launching the host, so hooks and commands share the environment and write access. Use separate worktrees when concurrent tasks change code.
# Set before launching the host; use your actual repository path.
export REFARIO_DATA_DIR="/path/to/repo/.refario/pilot-01"
# Inside the host session, use its native session ID:
node plugins/refario-agent-performance/scripts/start.mjs --repository "/path/to/repo" --task "Describe your task" --session-id "NATIVE_SESSION_ID"Use the session ID and host from the SessionStart hook; set REFARIO_AGENT_HOST to codex or claude-code as appropriate. Without native identity, replace --session-id with --manual and label native tool evidence unavailable. Repeat a start using its --capture-id.
Verify first. Record your decision.
Choose an existing npm script that checks the task objectively. After reviewing the result, explicitly accept, reject, or partially accept it. Passing tests alone are not acceptance.
node plugins/refario-agent-performance/scripts/review.mjs --repository "/path/to/repo" --task "Describe your task" --capture-id "CAPTURE_ID_FROM_START" --verify-script test --accept yes --no-uploadReplace test with a suitable existing npm script and yes with your actual decision: yes, no, or partial. Omit --verify-script if none is suitable; the result will remain unverified. Wait for a human decision before recording acceptance. Armed v0.1.2 captures remain pending through session end and resume until reviewed.
Local evidence, explicit limits.
Records default to ~/.refario/sessions. Model and reasoning can be missing; active time means start to first stop, with later rework recorded separately. Manual fallback capture does not provide native tool events. Compare similar tasks manually and keep monetary cost unavailable without authoritative evidence.
Existing customer economics remain available.
Platform accounts, telemetry, and customer margin are separate from installing this plugin.
Make the next task a useful experiment.
Capture the evidence, verify the result, and decide what to try next.