run-evals

name	run-evals
description	Run evaluations for one, multiple, or all skills using the agent orchestration framework. Make sure to use this skill whenever the user asks to run evals, test a skill's performance, run benchmarks, or compare baseline versus with-skill execution.
metadata	{"internal":true}

Run Skill Evals

1. Read Framework: Read <target-package-root>/evals/README.md for understanding the difference between per-skill evals and cross-skill evals (where <target-package-root> is the directory containing the .agents or skills folder).
2. Locate Targets: Find target evals/evals.json files inside .agents/skills/ and/or skills/. For cross-skill evaluations, look for *_evals.json files directly in <target-package-root>/evals/.
3. Determine Agent Configuration: Check the agent_config field in the target target JSON file to determine the environment/harness to spawn. If agent_config is "bare-agent", spawn a subagent with the bare-agent profile. If it is a specific contributor profile (e.g. "reidbaker-agent"), use that profile to provide the necessary contributor context.
4. Orchestrate: By default, run an Integration Test by spawning a single With-Skill subagent using Workspace: branch and the identified agent_config.
   • Provide the task prompt. See resources/with_skill_execution_prompt.md for the template. When filling in <path-to-skill>, you MUST use a relative path from the repository root, not an absolute path. If you are running a cross-skill evaluation, fill in <path-to-skill> with "none (cross-skill meta-eval)". Also, replace <target-package-root> with the actual directory path in both templates.
   • Only if the user explicitly requests a comparison or benchmark, also spawn a Baseline subagent. See resources/baseline_execution_prompt.md for the template. Instruct the subagent(s) to return their git diff and verification outputs (dart pub get, dart format, dart analyze, dart test) without committing. Ensure you instruct them to run these commands exclusively from within the <target-package-root> directory to avoid analyzing unrelated packages. CRITICAL: You must explicitly warn the subagent(s) to confine all file edits strictly to their current working directory and avoid using absolute paths to modify the parent workspace. WORKSPACE LIMITATION WARNING: If the user has multiple active workspaces mounted, the Workspace: branch feature will fail. In this situation, you MUST warn the user that running concurrent evaluations in Workspace: inherit mode will cause git state bleed and cross-eval pollution (e.g., changes made by a failure scenario will be visible to a success scenario running simultaneously in the same shared directory). Instruct the user to fix this by closing all workspaces except the primary package workspace, and then re-run the evaluations. Do NOT silently fallback to Workspace: inherit for concurrent tasks.
5. Grade: Parse the combined rubric (resolving repo_criteria + evals.json expectations). Use the grading instructions in resources/agent_judge_prompt.md. When an expectation fails, you MUST explicitly list both the expectation and what was actually found that caused the failure.
6. Artifact: Grade the outputs and generate a Markdown artifact (e.g., <skill>_eval_results.md) containing the metadata, pass/fail rationale, and raw diffs/stdout.