The Agent Skills Directory

[SAFE]: The skill defines a structured workflow for AI-assisted coding using the KAIROS protocol. All findings are consistent with the intended functionality of a developer-oriented coding assistant.
[COMMAND_EXECUTION]: The skill identifies and executes build, test, and deployment commands (e.g., npm run dev:test) and search commands (e.g., rg) found within the project's README.md. This represents a surface for indirect prompt injection. 1. Ingestion points: README.md in the project root. 2. Boundary markers: Absent. 3. Capability inventory: git, npm, rg, stat, and tail subprocess calls across SKILL.md and KAIROS.md. 4. Sanitization: Absent, although the skill logic requires the agent to ask for human confirmation if documented commands fail or require modification.
[PROMPT_INJECTION]: The skill uses strict instructional language (e.g., "Every code change MUST follow") and the must_obey parameter to enforce adherence to the ELITE AI CODING STANDARDS. These are internal logic constraints for the agent's operation and do not target the system prompt.
[DATA_EXFILTRATION]: Local protocol definitions and execution proofs (nonces, hashes) are transmitted to the configured KAIROS MCP server via tools like kairos_mint and kairos_attest. This data transfer is necessary for the state-machine-based workflow orchestration provided by the author.

kairos-code