The Agent Skills Directory

[COMMAND_EXECUTION]: The skill is authorized to execute development, self-test, and acceptance commands locally. This capability is inherent to its primary purpose as an autonomous developer agent.
[DATA_EXFILTRATION]: While the skill has broad read access to the local project files, the instructions explicitly forbid sending external messages, pushing to remote repositories, or interacting with production data, mitigating the risk of data theft.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it ingests project-level documentation to define its behavior and requirements.
Ingestion points: Reads project-specific guidance from AGENTS.md, CLAUDE.md, PROJECT_RULES.md, and ROADMAP.md.
Boundary markers: Implements a sequential stage-gate process where a primary agent must verify the output of each sub-agent stage before proceeding.
Capability inventory: Possesses file write permissions and shell command execution via the project-execution, project-review, and project-acceptance skills.
Sanitization: The skill relies on agent-to-agent verification of outputs but lacks explicit sanitization of input documents.

project-autopilot