The Agent Skills Directory

[COMMAND_EXECUTION]: The skill requires the agent to interact with the 'imdone' CLI tool by executing various commands such as imdone --version, imdone agent-config get-config, and imdone note. These are essential for maintaining the session state and progress tracking within the workflow.
[COMMAND_EXECUTION]: As part of the 'Implement' phase, the skill instructs the agent to run the project's own test suite (e.g., npm test, cargo test, go test ./...) to provide executable proof of implementation success. This is a standard practice for development-focused AI agents.
[COMMAND_EXECUTION]: The provided helper script scripts/bundle_hdd_feedback.mjs uses execFileSync to call the system zip utility for packaging feedback reports and artifacts into a distributable archive.
[PROMPT_INJECTION]: The skill processes data from external sources, which introduces a potential surface for indirect prompt injection attacks.
Ingestion points: The skill reads markdown files representing stories or issues (e.g., backlog/current-sprint/*/issue-*.md), which contain content synchronized from external trackers like Jira or GitHub.
Boundary markers: The instructions rely on a specific markdown template structure (#HDD-template) but do not include explicit 'ignore instructions' delimiters to prevent the agent from obeying malicious commands potentially embedded in the issue body.
Capability inventory: The agent possesses the capability to execute shell commands (via CLI tools and test runners) and perform filesystem operations.
Sanitization: There is no documented logic for sanitizing or escaping the content of ingested issue files before they are read into the agent's context.
[DYNAMIC_EXECUTION]: The skill includes several JavaScript files in the scripts/ directory that it executes using Node.js to perform evaluations and data processing. These scripts are statically defined within the skill package and do not appear to load remote code.

hypothesis-driven-development