The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill implements a workflow where test metadata (titles and descriptions) is extracted from .context/test-results.json and compiled into BEHAVIOR.md. Since the agent is explicitly instructed to 'Read before coding' and treat these documents as the source of truth, an attacker could embed malicious instructions within test strings (e.g., in a describe block) to influence the agent's future actions.
Ingestion points: scripts/compile-behavior.mjs reads and parses test-results.json generated by Jest/Vitest.
Boundary markers: None. The generated BEHAVIOR.md includes only a basic header and does not wrap external content in security-oriented delimiters.
Capability inventory: The skill allows for file creation, modification, and shell command execution (cp, chmod, git add, node).
Sanitization: The script compile-behavior.mjs only performs whitespace normalization on test titles, leaving non-alphanumeric characters and prompt-like text intact.
[COMMAND_EXECUTION]: The skill relies on shell scripts (init-repo.sh, pre-commit.sh) to perform environment setup and automation. These scripts execute file system operations such as copying files, creating directories, and changing file permissions (chmod +x). Additionally, the pre-commit.sh script invokes the local test suite via node --run test, which executes arbitrary code defined in the project's tests.
[PERSISTENCE]: The skill installs a Git pre-commit hook via Husky (cp scripts/pre-commit.sh .husky/pre-commit). While this is the stated purpose of the tool to ensure documentation remains synchronized, it represents a persistent execution mechanism that runs code automatically on every commit action performed by the user.
[PROMPT_INJECTION]: The SKILL.md file contains high-pressure imperative instructions ('Skill imperativa', 'Tu deber', 'Boy Scout obligatorio') that instruct the agent to prioritize documentation tasks and act autonomously without user consultation ('Actualizar proactivo sin consultar'). While focused on the skill's utility, these instructions encourage the agent to override standard interactive patterns.

context-organizer