The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and act upon data from external, potentially untrusted sources including GitHub/GitLab issue bodies, technical specifications, and project-level configuration files (CLAUDE.md, AGENTS.md).
Ingestion points: Identified in SKILL.md and references/phase-library.md (processing issue lists, task requirements, and spec documents).
Boundary markers: The skill does not define clear delimiters or specialized prompts to neutralize potential instructions embedded within the processed natural language data.
Capability inventory: The agent has broad permissions to execute shell commands (git, gh-cli), modify the filesystem, and interact with the web via agent-browser.
Sanitization: There is no evidence of filtering or sanitizing external content before it is interpolated into the agent's internal reasoning.
[DYNAMIC_EXECUTION]: The skill implements a self-improvement loop in references/learning-and-audit.md, allowing the agent to update its own instructions (skills and reference files) based on task outcomes. This self-modification capability represents a mechanism where generated instructions are subsequently executed.
[COMMAND_EXECUTION]: The skill utilizes shell-based tooling for core operations, specifically invoking git for repository management and gh-cli for task lifecycle management. These commands are executed as part of the normal operation of the autopilot workflow.

autopilot