The Agent Skills Directory

[PERSISTENCE]: The skill implements a 'continual learning' system through the agents-memory-updater persona defined in references/agents/agents-memory-updater.md. This agent is instructed to mine transcript deltas and update the CLAUDE.md and AGENTS.md files with learned preferences and facts. Since these files are defined in references/repo-orientation.md as top-priority instruction files, this allows the agent to permanently modify its own future system-level instructions based on conversation history.
[EXTERNAL_DOWNLOADS]: The SKILL.md and several reference files (e.g., references/codebase-architecture-language.md, references/interface-design.md) contain instructions to install external skills using the scripts/install-external-skills.py script. This facilitates the download and integration of external code into the agent's environment.
[INDIRECT_PROMPT_INJECTION]: The memory-updating mechanism creates a vulnerability surface for indirect prompt injection. Malicious instructions embedded in data processed by the agent (such as PR comments or documentation) that appear in transcripts could be identified by the agents-memory-updater as 'durable preferences' and written to the agent's permanent instruction files.
[COMMAND_EXECUTION]: Several utility scripts included in the skill, such as scripts/repo_scan.py and scripts/architecture_diagram_generator.py, utilize subprocess.run to execute shell commands like git and rg (ripgrep) on the local filesystem.

architecture