The Agent Skills Directory

[PROMPT_INJECTION]: Indirect Prompt Injection Surface Area
Ingestion points: The orchestration strategies described in strategies.md and examples.md involve tools like web-doc-resolver and codebase-analyzer which ingest data from potentially untrusted external sources.
Boundary markers: The instructions do not specify or require the use of delimiters (e.g., XML tags or triple quotes) or explicit 'ignore instructions' warnings when processing data from these sources.
Capability inventory: The workflow involves agents with significant capabilities, such as feature-implementer (which writes code to the filesystem) and loop-agent (which executes commands via cargo and local shell scripts).
Sanitization: There is no provision or guidance for sanitizing external data before it is interpolated into prompts or used to drive implementation decisions.
[COMMAND_EXECUTION]: Automated Command and Script Execution
The skill instructs the agent to perform autonomous execution of local commands and scripts as part of quality gates and iterative loops, including cargo test, cargo clippy, and ./scripts/code-quality.sh (as seen in examples.md and quality-gates.md). While standard for software development, these automated executions are triggered by the coordination logic.

agent-coordination