Security: do not auto-discover adapters from untrusted eval bundles

[ComicBit]

Summary

The bmad-eval-runner scripts automatically load adapter.json or .bmad-eval-adapter.json from beside the cases/queries file. An adapter controls the executable argv and optional host-environment forwarding, and the runner passes it to subprocess.run.

A developer who runs an untrusted evaluation bundle can therefore execute arbitrary local commands as the evaluator user, with any environment values requested by the bundle's adapter. The same root cause exists independently in both run_evals.py and run_triggers.py.

Affected paths

• skills/bmad-eval-runner/scripts/run_evals.py
• skills/bmad-eval-runner/scripts/run_triggers.py
• skills/bmad-eval-runner/references/platform-adapter.md
• skills/bmad-eval-runner/SKILL.md

Proposed fix

• Remove sibling adapter-file auto-discovery.
• Keep only explicit --adapter and BMAD_EVAL_ADAPTER configuration paths.
• Centralize adapter resolution so both runners share the same trust-boundary behavior.
• Add regression tests for sibling-file rejection and trusted explicit configuration.

Impact

This is untrusted-file-assisted arbitrary command execution. It requires a developer to run the bundle; it is not a network-exposed or zero-click service.

[coderabbitai]

🔗 Related PRs

#32 - Replace bmad-init dependency with direct config loading [merged]
#101 - fix(eval-runner): harden fixture isolation with parent-held digests [open]

---

📝 Issue Planner

_{Check the box below or use the @coderabbitai plan command to generate an implementation plan and prompts that you can use with your favorite coding assistant.}

• Create Plan

---

🧪 Issue enrichment is currently in open beta.

You can configure auto-planning by selecting labels in the issue_enrichment configuration.

To disable automatic issue enrichment, add the following to your .coderabbit.yaml:

issue_enrichment:
  auto_enrich:
    enabled: false

💬 Have feedback or questions? Drop into our discord!