The Agent Skills Directory

[Command Execution]: The skill executes various shell commands and test runners as part of its development workflow.
Evidence: Step 6.5.2 executes npm test (or yarn/pnpm equivalents) to validate project scaffolding.
Evidence: Step 6.6.1 executes discovered API test scripts using bash ./api-test.sh.
[Dynamic Execution]: The skill executes local scripts that are discovered at runtime within the project structure.
Evidence: Step 6.6.1 runs scripts like api-test.sh, api-check.sh, etc., that were identified in Step 5.5. Executing scripts found in the workspace is a potential vector for code execution if the project contents are compromised or malicious.
[External Downloads]: The skill may attempt to install software packages during automated remediation steps.
Evidence: Step 6.5.2 instructs the agent to install missing dev dependencies if the test script fails. This involves downloading and installing external packages based on runtime error analysis.
[Indirect Prompt Injection Surface]: The skill ingests and acts upon information from multiple external project files without explicit sanitization or boundary markers.
Ingestion points: Multiple files including feature spec, plan.md, constitution.md, and discovered test scripts like api-test.sh (Step 2 and Step 5.5).
Boundary markers: No specific delimiters or instructions to ignore embedded commands are used when these files are read into the agent's context.
Capability inventory: The skill possesses the ability to modify the file system, execute shell commands, and perform network probes via curl (Step 6.6.2).
Sanitization: No explicit sanitization or validation of the ingested content is documented. Malicious instructions placed in these artifacts could potentially influence the agent's behavior during the implementation process.

implementing-code