The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill acts as a surface for indirect prompt injection by processing untrusted user feature requests and feature descriptions to drive its planning and implementation workflow.
Ingestion points: User requests for feature development (e.g., "implement X feature") identified as auto-invoke triggers in SKILL.md.
Boundary markers: The skill employs structured markdown templates (e.g., implementation-plan-template.md) to guide agent responses but lacks explicit "ignore embedded instructions" delimiters for data ingested from the user or existing codebase.
Capability inventory: The skill uses glob and grep for file search, TodoWrite for task tracking, Task for sub-agent coordination, and shell execution to run local validation scripts (validate-increment.sh) that invoke npm commands.
Sanitization: No explicit sanitization or filtering logic is present to validate user input before it influences the generated implementation plans or commands.
[COMMAND_EXECUTION]: The skill relies on local shell scripts and standard development tools to enforce its research-backed workflow.
The instructions in SKILL.md and REFERENCE.md guide the agent to execute ./scripts/validate-increment.sh (or .bat) and ./scripts/check-file-size.sh.
These scripts execute standard project-level commands such as npm run lint, npm run type-check, and npm test to verify code quality and functionality between increments.

feature-orchestrator