The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute several shell commands for auditing the repository state and measuring file sizes. Evidence includes the use of wc -l, find, du -sh, rg (ripgrep), and git diff within SKILL.md (Phases 0, 1, and 4).
[PROMPT_INJECTION]: The skill presents a surface for indirect prompt injection through its "graduation" mechanism. It encourages the agent to extract knowledge from external memory files (such as those in ~/.claude/projects/) and promote them to project-level instructions (CLAUDE.md, AGENTS.md).
Ingestion points: The skill reads from docs/, .codestable/, and external agent memory directories (~/.claude/, ~/.codex/, ~/.config/opencode/).
Boundary markers: There are no explicit instructions to wrap or sanitize content being promoted from memory files into project rules.
Capability inventory: The skill has the capability to write to and modify critical agent instruction files (CLAUDE.md, AGENTS.md) and project documentation.
Sanitization: No explicit sanitization or validation of the content being moved between files is defined in the instructions.

cs-docs-neat