The Agent Skills Directory

[CREDENTIALS_UNSAFE]: Hardcoded credentials detected in the documentation examples. Unlike other sections using 'xxxxx' placeholders, the 'Mixed/Informal' format section contains high-entropy strings that appear to be actual credentials:
Reddit client ID: Y1FgKALKmb6f6UxFtyMXfA
Reddit secret: -QLoYdxMqOJkYrgk5KeGPa6Ps6vIiQ
[CREDENTIALS_UNSAFE]: The skill explicitly targets and encourages the discovery of centralized sensitive files containing secrets, including:
~/Documents/Access.txt
~/.secrets/keys.txt
~/.credentials.txt
[DATA_EXFILTRATION]: The skill instructs the agent to transmit extracted secrets to various external APIs (OpenAI, Anthropic, Render, Reddit, Replicate) using curl. While described as a validation step, this pattern involves sending raw credentials over the network to external endpoints.
[COMMAND_EXECUTION]: The skill utilizes shell commands to perform its operations, which involves executing network requests with extracted secrets and writing to the local filesystem:
curl for key validation
cat > .env for writing environment files
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection as it ingests untrusted data from user-specified files and processes it to perform sensitive operations without sanitization.
Ingestion points: Files like ~/Documents/Access.txt (SKILL.md)
Boundary markers: None provided; the skill reads and parses the entire file content.
Capability inventory: curl for network requests, file writing for .env creation, and jq for JSON processing.
Sanitization: No evidence of input sanitization or validation of the file content before processing.

credentials