The Agent Skills Directory

[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection because it ingests and summarizes untrusted data from the target codebase to generate documentation.
Ingestion points: The skill performs a full recursive census and reads the contents of all source files (e.g., package.json, Cargo.toml, .ts, .py, .rs) to generate summaries, architecture overviews, and pattern documentation as described in references/generate-workflow.md.
Capability inventory: The agent has permissions to execute shell commands (git, ls) and perform extensive file system writes across multiple directories.
Boundary markers: The instructions do not provide delimiters or "ignore" directives to prevent the agent from following malicious prompts embedded within the source code files it is documenting.
Sanitization: The discovery and generation workflows lack explicit sanitization or validation of the contents retrieved from the repository before processing them.
[COMMAND_EXECUTION]: The skill instructs the agent to execute shell commands to interact with the environment and analyze the codebase.
The "Companion check" in SKILL.md uses the ls command to search for installed extensions in platform-specific paths like ~/.claude/skills/, ~/.agent/skills/, and within the local project directory.
The update and generate workflows rely on git commands (e.g., git diff, git log, git ls-files) to identify project boundaries, compute coverage, and track changes between commits.
[EXTERNAL_DOWNLOADS]: The documentation encourages the use of npx to download and install the skill from the vendor's repository (AbsolutelySkilled/AbsolutelySkilled). This is a standard installation method for the platform but involves executing remote code.

codedocs