The Agent Skills Directory

[COMMAND_EXECUTION]: The afk-cook script executes the AI agent with the --permission-mode bypassPermissions flag. This configuration enables the agent to perform actions such as executing shell commands, reading or writing files, and utilizing the GitHub CLI (gh) without individual user prompts. This high level of autonomy is required for the skill's primary function of unattended batch processing but requires users to trust the source of the tasks and issues being executed.
[INDIRECT_PROMPT_INJECTION]: The skill processes data from GitHub issues to define tasks. Since the agent reads issue bodies and comments via gh issue view, there is a risk that instructions embedded within an issue could be followed by the agent, potentially leading to unauthorized actions during the unattended run.
Ingestion points: Issue content is retrieved in slice-prompt.md and dependency markers are parsed from issue bodies in the afk-cook script.
Boundary markers: The prompt template includes task scope instructions ('Work only on this single issue') but does not employ specific delimiters or warnings to ignore instructions embedded within the issue data.
Capability inventory: Full autonomous shell access and repository management via pre-authenticated tools.
Sanitization: External issue text is processed as raw data without explicit sanitization or filtering.
[DYNAMIC_CONTEXT_INJECTION]: The slice-prompt.md template includes the syntax !`git log -n 5 --oneline`. This feature executes shell commands at the time of prompt generation to provide the agent with project context, specifically the most recent git history.
[EXTERNAL_DOWNLOADS]: The documentation references installation using npx skills@latest add GuillermoMurillo/memo-flow, which fetches the runner script and templates from the author's public repository.

afk-cook