The Agent Skills Directory

[PROMPT_INJECTION]: The skill utilizes and teaches the Authority and Commitment principles to enforce compliance with TDD guidelines. It employs imperative and non-negotiable language such as YOU MUST, No exceptions, and Violating the letter of the rules is violating the spirit, which are specifically designed to override the agent's natural reasoning and potential rationalizations during task execution.
[COMMAND_EXECUTION]: The provided render-graphs.js utility script uses the Node.js child_process.execSync module to execute the system dot command (Graphviz) for rendering diagrams. This is a local command execution pattern provided as a developer utility for visualizing skill workflows.
[INDIRECT_PROMPT_INJECTION]: The skill framework involves testing and documenting agent behavior by ingesting subagent outputs in pressure scenarios, creating an attack surface for indirect instructions. (1) Ingestion points: Subagent rationalizations and choices captured verbatim during the RED phase of testing as described in testing-skills-with-subagents.md. (2) Boundary markers: No explicit markers or delimiters are defined for the ingested subagent content. (3) Capability inventory: Ability to write new skill files (SKILL.md) and execute shell scripts (render-graphs.js). (4) Sanitization: The methodology explicitly requires documenting failures verbatim, with no sanitization or filtering of the captured subagent data before interpolation into documentation.

writing-skills