The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill collects arbitrary text from user messages, such as tasks, requests, and consultation questions, and interpolates them into markdown templates used to prompt a secondary coding agent (alcode).
Ingestion points: User messages triaged in references/channel-handling.md, request capture in references/working-session.md, and questions in references/runbooks/consultation.md.
Boundary markers: Relies on simple text labels (e.g., 'Task:', 'Request:') within markdown files to delimit user-provided content.
Capability inventory: Uses the exec tool to perform sensitive operations including filesystem writes, repository management via git, and executing the alcode coding agent which can modify the codebase.
Sanitization: User-provided text and resource URLs are preserved verbatim without sanitization or escaping.
[EXTERNAL_DOWNLOADS]: The skill allows cloning repositories from arbitrary, user-provided URLs as part of the onboarding process.
Evidence: The references/runbooks/project-lifecycle.md file defines a procedure to clone a repository URL provided by the user into the project environment.
[COMMAND_EXECUTION]: The skill relies on the execution of multiple local CLI tools to perform its intended functions.
Evidence: Extensive use of the exec tool to run git, alproject, alcode, alignfirst, and openclaw commands is documented across all instruction files.

alignfirst-openclaw-playbook