The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill implements a 'lane' system for cross-runtime dispatch, allowing it to spawn external CLI tools as child processes based on configurations found in mission.yml and references/lanes.md. This enables the execution of arbitrary binaries defined in the workflow policy.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external sources to drive its logic.
Ingestion points: The skill pulls mission briefs, goals, and step instructions directly from issue summaries, descriptions, and comments on platforms like GitHub, GitLab, Jira, and Linear (references/tracker-integration.md).
Boundary markers: The system uses a structured brief.md format and includes a safety constraint in references/tracker-integration.md §Safety Constraints #5 that instructs the agent to treat issue content as data rather than instructions.
Capability inventory: The orchestrator has the capability to write to the file system, manage git worktrees, perform git commits, and post comments/labels back to the external tracker via MCP servers or CLI tools.
Sanitization: While the skill documentation claims to treat external messages as data, the primary workflow involves an LLM interpreting this untrusted text to structure the 'Mission Brief' which governs subsequent automated actions.
[COMMAND_EXECUTION]: The executor performs extensive command-line operations, including complex git worktree management, commit authorship validation, and the invocation of tracker CLI tools (gh, glab, linear, jira) as detailed in references/executor.md and references/tracker-integration.md.
[EXTERNAL_DOWNLOADS]: The skill connects to and fetches content from well-known project management and version control services (GitHub, GitLab, Atlassian/Jira, Linear) to synchronize state and retrieve task instructions.

factory-mission