The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to fetch and process user-generated content from external app stores, which acts as an untrusted input surface.
Ingestion points: Untrusted data enters the agent context via the /appstore/reviews, /googleplay/reviews, and /appstore/app (description) endpoints as documented in reference/endpoints.md and SKILL.md.
Boundary markers: The instructions lack explicit delimitation or "ignore instructions" wrappers for the external review text ingested by the agent.
Capability inventory: The skill provides network access via scripts/crawlora.sh to a restricted set of vendor API routes. It has no arbitrary file-write, command execution, or evaluation capabilities.
Sanitization: The helper script scripts/crawlora.sh implements defensive input handling, including query parameter encoding and standard input streaming for POST bodies, to prevent injection into the underlying shell command.
[SAFE]: The skill's implementation of scripts/crawlora.sh follows security best practices to mitigate common attack vectors:
Credential Protection: The CRAWLORA_API_KEY is passed to curl using a temporary configuration file with restricted permissions (chmod 600), preventing the secret from appearing in process lists (e.g., ps output).
Input Validation: The script uses a strict whitelist and regular expression matching for API paths, and it explicitly rejects suspicious characters (like .., //, ?, or #) that could be used for directory traversal or request smuggling.
Local File Protection: The script rejects the @ character in query parameters to prevent curl from inadvertently disclosing local system files.

app-market-opportunity-research