The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill guides the generation of new agent instructions and scripts based on user input gathered during an interview phase. This workflow creates a surface for indirect prompt injection if a user provides malicious instructions that the agent then incorporates into the drafted skill content. * Ingestion points: User responses during the Phase 1 Interview (described in references/create.md) are used to synthesize skill drafts. * Boundary markers: The skill employs a structured five-phase development lifecycle and a mandatory Phase 5 Review checklist to validate output quality and safety. * Capability inventory: The agent is instructed to create and modify files, including SKILL.md, Python scripts, and Markdown references, within the local workspace. * Sanitization: The instructions include 'self-critique loops' and 'absolute bans' lists to help the agent identify and remove problematic patterns during the drafting process.- [COMMAND_EXECUTION]: The instructions direct the agent to utilize common developer tools such as grep for searching project history and ruff for Python formatting and linting. These operations are appropriate for the intended development-focused use case.- [EXTERNAL_DOWNLOADS]: The skill references documentation hosted at agentskills.io and suggests using the uv tool to manage the ruamel.yaml dependency for specific scripting tasks. These are documented neutrally as standard project dependencies.

skill-authoring