The Agent Skills Directory

[SAFE]: The skill is purely instructional and serves as an organizational entry point for other tools. It does not perform any high-risk operations.
[DATA_EXPOSURE]: The skill reads specific project files within the .codestable/ directory to understand context. This is limited to project metadata and documentation (e.g., attention.md, system-overview.md) and does not target sensitive system credentials, SSH keys, or environment secrets.
[PROMPT_INJECTION]: The instructions are clear and follow standard operational logic. There are no attempts to bypass safety filters, extract system prompts, or override agent constraints.
[COMMAND_EXECUTION]: The skill uses local file system discovery (Glob) and reading (Read) to perform its routing task. It does not execute arbitrary shell commands or system-level scripts.
[INDIRECT_PROMPT_INJECTION]: While the skill ingests content from external project files which could theoretically contain malicious instructions, the impact is minimized because this skill only provides routing advice to the user and does not execute the file content as code or commands.
Ingestion points: .codestable/attention.md, .codestable/reference/system-overview.md.
Boundary markers: Absent; the skill reads the files directly to gain project context.
Capability inventory: Reading local files and directory structures via Glob and Read tools.
Sanitization: None; the skill relies on the LLM to interpret the text for routing purposes.

cs