The Agent Skills Directory

[COMMAND_EXECUTION]: The skill's workflow definitions (YAML 'pieces') utilize the Bash tool to perform system-level operations required for software development, such as running tests (npm test), building projects (npm run build), and executing infrastructure plans (terraform plan). These are primarily used in pieces like backend.yaml, terraform.yaml, and takt-default.yaml.
[EXTERNAL_DOWNLOADS]: The skill enables WebSearch and WebFetch tools to retrieve information from the internet for research tasks (e.g., in research.yaml and deep-research.yaml). It also uses the GitHub CLI (gh) to download Pull Request and Issue data from GitHub repositories, as described in the gather-review.md facet.
[PROMPT_INJECTION]:
Indirect Prompt Injection Surface: The skill ingests untrusted data from external web pages and GitHub Issues/PRs (e.g., in research-dig.md and gather-review.md). This external content is interpolated into the agent context. The agent possesses high-privilege capabilities including Bash execution and file Edit/Write permissions (defined in piece files like backend.yaml and terraform.yaml). No explicit sanitization or 'ignore embedded instructions' delimiters are applied to this external content before processing.
Operational Behavioral Steering: The skill includes specific instructions in ai-fix.md ("あなたの「修正済み」という認識が間違っています" / "Your recognition of 'fixed' is wrong") designed to override the agent's internal assumptions and correct common hallucination patterns. These are benign operational directives intended to improve reliability rather than malicious overrides.

takt-facet-builder