tommy-ca/grok-build-plugins/check-agent-compatibility

Check agent compatibility

Read `../../HARNESS.md` first.

Trigger

Use when the user wants the full compatibility pass for a repo.

Workflow

Grok MAX_SUBAGENT_DEPTH is 1. This skill runs in the parent. Fan out four children in one message. Each spawn_subagent needs prompt, description (3-5 words), subagent_type, background: true (TUI default is false), and isolation: "none" unless HARNESS says otherwise. Omit model to inherit. Do not send effort on spawn. Children do not spawn.

1. spawn_subagent subagent_type agent-compatibility:compatibility-scan-review to run the CLI and capture the raw repository score.
2. spawn_subagent subagent_type agent-compatibility:startup-review to verify whether the repo can be booted by an agent.
3. spawn_subagent subagent_type agent-compatibility:validation-review to check whether an agent can verify a small change without an unnecessarily heavy loop.
4. spawn_subagent subagent_type agent-compatibility:docs-reliability-review to see whether the documented setup and run paths match reality.
5. Use one subagent per task. Do not collapse these checks into one agent prompt.
6. Join with get_command_or_subagent_output using task_ids and a positive timeout_ms.
7. Compute an internal workflow score as the rounded average of Startup, Validation Loop, and Docs Reliability scores.
8. Compute an Agent Compatibility Score as round((deterministic_score * 0.7) + (workflow_score * 0.3)).
9. Synthesize the results into one final response.

When scoring internally, use specific non-round workflow scores for the behavioral checks rather than coarse round buckets. If startup, validation, or docs mostly work, treat them as good-with-friction rather than defaulting to the mid-60s. Do not create a low workflow score just because logs are noisy or the error text is rough.

If the deterministic scanner cannot be run because of tool environment issues, say that separately and do not treat it as a repo defect or penalize the repo.

Output

Respond in markdown, but keep it minimal. Do not use fenced code blocks.

Show only one score, as a level-two heading. ## Agent Compatibility Score: N/100. Do not show how it was computed unless the user explicitly asks for a breakdown.

Then a flat, prioritized list labeled Top fixes with one issue per line, each line starting with - .

Fold deterministic and behavioral findings into that one list. Focus on the fixes that would most improve real agent workflows.