The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to use standard shell commands such as grep to perform search operations across the codebase for animation-related keywords and git rev-parse to identify the current commit hash for plan metadata. These are routine development operations.- [INDIRECT_PROMPT_INJECTION]: As the skill is designed to read and process external repository content, it is inherently exposed to indirect prompt injection from malicious code or comments. The skill proactively mitigates this risk through 'Hard Rule 4', which explicitly directs the agent to treat all file contents as inert data and to flag any embedded instructions intended to steer its behavior.
Ingestion points: The agent reads source code and project metadata during the Recon and Audit phases (SKILL.md).
Boundary markers: The skill contains a specific rule instructing the agent to ignore directives within files and treat them as data only.
Capability inventory: The skill can write plan files to a specific /plans directory and has the ability to dispatch subagents for analysis tasks.
Sanitization: The workflow includes a mandatory 'Vet' phase (Phase 3) where the agent must manually confirm findings against the source code before generating plans, reducing the risk of automated obedience to injected instructions.

improve-animations