The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill operates by having the agent read, modify, and verify codebases, which inherently creates an attack surface for indirect prompt injection. Malicious instructions placed in source code comments or documentation within the target project could potentially influence the agent's actions during the implementation cycle.
Ingestion points: Project source files and documentation processed during the "Increment Cycle" defined in SKILL.md.
Boundary markers: SKILL.md includes "Rule 0.5: Scope Discipline" and "Rule 1: One Thing at a Time" to constrain agent behavior, though these are guidelines rather than technical boundaries.
Capability inventory: The skill utilizes file-writing capabilities, git commit operations, and shell command execution for testing and building (SKILL.md).
Sanitization: There is no evidence of automated sanitization for content ingested from the codebase before it is processed by the agent.
[COMMAND_EXECUTION]: The skill explicitly directs the agent to execute repository-specific commands for verification, such as npm test, ./gradlew test, pytest, and build tools. It also suggests using shell commands like grep for secret detection in references/security-checklist.md. These commands are standard for the development workflow described.
[EXTERNAL_DOWNLOADS]: The reference documentation mentions several well-known development tools and libraries, including axe-core, pa11y, lighthouse, and bundle analyzers executed via npx (references/accessibility-checklist.md, references/performance-checklist.md). These are official tools from the NPM registry and are documented as part of standard engineering practices.

incremental-implementation