The Agent Skills Directory

[DATA_EXPOSURE]: The skill's authentication client in scripts/tapd_client.py performs broad scans for authentication tokens in sensitive user files beyond the skill's directory.
Evidence: The _load_tapd_token function searches for TAPD_ACCESS_TOKEN in ~/.zshrc, ~/.env, and several other locations in the user's home directory to automate login.
[INDIRECT_PROMPT_INJECTION]: The skill fetches and processes user-generated content (requirement and bug descriptions) from the TAPD platform, which serves as an attack surface for indirect prompt injection.
Ingestion points: The skill fetches external data via the get_stories_or_tasks and get_bug commands in scripts/tapd.py.
Boundary markers: The skill does not use specific delimiters to wrap external content when presenting it to the agent.
Capability inventory: The agent has the ability to execute shell commands, write files, and perform network operations.
Sanitization: No sanitization or filtering of instructional content is performed on the data fetched from the external TAPD platform.
[COMMAND_EXECUTION]: The skill relies on shell command execution for all core operations and provides instructions for the agent to use pipes for data transformation.
Evidence: SKILL.md includes multiple examples where the agent is encouraged to pipe API results into python3 -c for dynamic JSON processing and reporting.
[DATA_EXFILTRATION]: The skill includes a specific command for sending data to an external webhook, which could be misused if the agent is manipulated.
Evidence: The send_qiwei_message command in scripts/tapd.py transmits markdown content to a URL defined by the BOT_URL environment variable.

tapd-idle