The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests and prioritizes project-specific instructions from .ai-factory/skill-context/aif/SKILL.md and project metadata files (e.g., package.json, composer.json). These files can contain instructions that override the agent's behavior.
Ingestion points: .ai-factory/skill-context/aif/SKILL.md, package.json, composer.json, and the content of skills installed from skills.sh.
Boundary markers: The skill mandates the use of an automated security-scan.py and a Level 2 manual semantic review of all external files.
Capability inventory: File system modification (Write, mkdir), network access (WebFetch), and shell command execution (Bash for node, python, and npx).
Sanitization: Automated detection of injection patterns via script and agent-led manual verification of instruction intent.
[EXTERNAL_DOWNLOADS]: The skill fetches and installs external agent skills from https://skills.sh using the npx skills command.
[COMMAND_EXECUTION]: The skill utilizes several executable scripts for its core logic, including a Node.js-based configuration patcher (update-config.mjs) and Python-based security utilities (security-scan.py, cleanup-blocked-skill.py). Execution is scoped within the allowed-tools configuration.

aif