The Agent Skills Directory

[COMMAND_EXECUTION]: The skill manages development processes and servers using bash and tmux sessions. It provides patterns for creating, reusing, and capturing output from these sessions to facilitate background task management.
[EXTERNAL_DOWNLOADS]: Recommends downloading standard .gitignore templates from the official GitHub repository (github/gitignore) during project initialization. This is a trusted source used for standard repository configuration.
[DATA_EXPOSURE]: Includes comprehensive repository hygiene rules that instruct the agent to identify and exclude sensitive files (e.g., .env, private keys) from version control using .gitignore.
[PROMPT_INJECTION]: The skill implements extensive behavioral guardrails and 'Always-On Rules' to ensure the agent follows a specific workflow and coding principles. These are standard instructional constraints.
[INDIRECT_PROMPT_INJECTION]: The skill has a surface for indirect prompt injection as it analyzes repository content that could be untrusted.
Ingestion points: Reads files, task/plan documents, and code comments in the investigation phase (SKILL.md, workflow.md).
Boundary markers: Employs a mandatory three-phase workflow (Investigate, Proposal, Implement) requiring human approval before any implementation occurs.
Capability inventory: Capable of writing to the filesystem (docs/task, docs/plan) and executing shell commands (delivery.md).
Sanitization: Relies on human review of proposals to mitigate risks from processed data.

pma