The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill interacts with external web pages and browser content, creating an attack surface for instructions embedded in the audited sites.\n
Ingestion points: The skill uses agent-browser to take snapshots and read console logs from portal.altschoolafrica.com and local origins in workflows/reproduce.md and workflows/exploratory-qa.md.\n
Boundary markers: SKILL.md and workflows/reproduce.md contain explicit warnings to treat browser data as untrusted and to ignore any 'run this command' instructions found on pages.\n
Capability inventory: The skill can navigate the browser, take screenshots, and execute local scripts (scripts/login.sh).\n
Sanitization: Instructions demand redaction of tokens, Authorization headers, and session bodies before logging evidence.\n- [DYNAMIC_EXECUTION]: scripts/login.sh generates and executes a Python script at runtime to parse local configuration.\n
Evidence: The script pipes a Python code block into python3 - to resolve preview origins from ~/.altschool-preview.json using the json and sys modules.\n- [COMMAND_EXECUTION]: The skill executes local system commands and specialized CLI tools to perform QA tasks and check prerequisites.\n
Evidence: scripts/preflight.sh and scripts/login.sh invoke agent-browser, op (1Password CLI), ss, lsof, and python3.\n- [CREDENTIALS_UNSAFE]: The skill accesses a local encryption key file to manage authentication for its browser driver.\n
Evidence: scripts/login.sh reads the contents of ~/.agent-browser/.encryption-key to export the AGENT_BROWSER_ENCRYPTION_KEY variable.\n
Context: The skill includes a preflight.sh script that verifies the key file has secure file permissions (mode 600 or 400).

altschool-portal-qa