The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to ingest and summarize data from untrusted external sources, including documentation, issue trackers, and remote repositories, during Stage 1 (Investigation). This information is recorded in the 'investigation-notes.md' artifact, which influences subsequent stages of design and implementation.
Ingestion points: investigation-notes.md.
Boundary markers: No specific delimiters or safety instructions are defined to separate untrusted research data from system prompts.
Capability inventory: File system modification, shell command execution (git, ripgrep, wc), and network operations via curl/wget.
Sanitization: The workflow relies on a human-in-the-loop verification gate in Stage 10 before finalization.
[REMOTE_CODE_EXECUTION]: Stage 1 (Investigation) explicitly allows the agent to run focused commands, reproductions, and small diagnostic scripts, including the inspection and cloning of upstream or sample repositories. While intended for legitimate engineering analysis, this represents a capability to execute code derived from external sources.
[COMMAND_EXECUTION]: The skill performs automated repository management tasks using shell commands, including 'git commit', 'git push', 'git merge', and 'git worktree' operations. These commands are executed during ticket bootstrap (Stage 0) and repository finalization (Stage 10).
[EXTERNAL_DOWNLOADS]: The workflow documentation and instructions permit the agent to download or clone external repositories, examples, and sample applications to verify integrations or establish current behavior during the investigation phase.

software-engineering-workflow-skill