The Agent Skills Directory

[SAFE]: The skill utilizes specific local directories for configuration and archiving (~/.config/makerskills/). This is standard behavior for the Maker Skills ecosystem to maintain user context and session history across updates.
[SAFE]: The skill includes an attack surface for indirect prompt injection via external research tools (deep-research, watch-video) and local project metadata. However, it incorporates robust grounding rules that mandate source attribution, prohibit fabricated quotes, and require explicit labeling of outputs as simulations, which effectively mitigates common injection risks.
Ingestion points: Untrusted data enters via external research tools and the portfolio.local.md configuration file.
Boundary markers: The protocol requires a mandatory "Simulated council" disclaimer and structured headers.
Capability inventory: Capabilities are limited to local file read/write within the skill's own configuration space and invocation of standard research/decision tools.
Sanitization: The agent is instructed to paraphrase documented positions and reason by analogy rather than blindly repeating input.
[SAFE]: No obfuscation, base64-encoded payloads, hardcoded secrets, or unauthorized remote code execution patterns were found in the skill's instructions or referenced dossiers.

maker-council