The Agent Skills Directory

[COMMAND_EXECUTION]: The scripts/atom_dashboard.py file implements a local HTTP server that acts as a development bridge. This bridge includes functionality to execute shell commands specified in the project's .atom-feature/manifest.json file. This capability is used to test the different 'surfaces' of a feature. Security is managed by using subprocess.run with shell=False to prevent command injection, and execution is explicitly disabled by default, requiring the user to pass an --allow-run flag when starting the dashboard.
[INDIRECT_PROMPT_INJECTION]: The skill ingests data from local project configuration files (JSON) and interpolates them into agent instructions and the dashboard UI. This creates a surface where manipulated project files could influence the agent's behavior. Mandatory Evidence Chain:
Ingestion points: Reads .atom-feature/manifest.json, acceptance.json, and presets.json from the target project directory.
Boundary markers: Absent; the skill relies on the structured nature of JSON data and the assumption that project files are within the user's local trust boundary.
Capability inventory: Includes command execution via subprocess.run, file writing for presets and status logs, and hosting a local web server.
Sanitization: Uses shell=False for executions, restricts the server to the loopback interface (127.0.0.1), enforces project-root path checks to prevent directory traversal, and implements a regex-based redaction system to mask secrets like API keys in the dashboard output.

lov-atom-feature-dev