The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes various system and network operations through a suite of included bash scripts to manage the development lifecycle.
Evidence: Uses the Bash tool to invoke scripts/add-finding.sh, scripts/file-feedback.sh, and scripts/lib/draft-pr.sh (SKILL.md).
Purpose: These scripts automate git commits, PR creation, and ticket status updates in Linear and GitHub using established CLI tools like git, gh, and linearis.
[INDIRECT_PROMPT_INJECTION]: The skill ingests technical plans from the local filesystem, which presents a surface for processing untrusted data if the repository is compromised.
Ingestion points: Technical plans are read from the thoughts/shared/plans/ directory (SKILL.md).
Boundary markers: The skill instructions do not define explicit delimiters to isolate plan data from the agent's instructions, though it operates within a controlled TDD phase structure.
Capability inventory: The agent has high-privilege access to tools such as Bash, Write, and Edit to implement the plans.
Sanitization: No specific sanitization or filtering is performed on the markdown plan content before the agent processes it.
[DATA_EXFILTRATION]: The skill transmits project metadata and code to external platforms (GitHub and Linear).
Evidence: Communication occurs with GitHub repositories and Linear team endpoints (scripts/file-feedback.sh).
Context: All operations target legitimate project resources, and the scripts specifically mention the author's own infrastructure (coalesce-labs/catalyst), indicating standard vendor-authorized behavior.

implement-plan