The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill audits the local project environment by reading file names, directory structures, and the content of go.mod via the setup-context.sh script. This information is used to guide the agent in generating scaffolding files. While necessary for the skill's function, reading local files represents an ingestion of untrusted data.\n
Ingestion points: setup-context.sh reads directory contents, the go.mod file, and git remote configuration from the local filesystem.\n
Boundary markers: The SKILL.md instructions explicitly direct the agent to propose diffs rather than overwriting existing files and to ask for user confirmation before proceeding with any changes.\n
Capability inventory: The skill uses a restricted Bash tool (setup-context.sh) to execute discovery commands and instructs the agent to generate file content for user review.\n
Sanitization: The setup-context.sh script implements a json_escape function to ensure data read from the filesystem is correctly formatted for the agent's context and does not break the JSON structure.\n- [COMMAND_EXECUTION]: The skill executes a pre-defined Bash script (setup-context.sh) to perform a read-only audit of the Go project. The script utilizes standard system tools such as go version, git remote, find, and grep to extract metadata about the project's structure and configuration.\n- [CREDENTIALS_UNSAFE]: The GitHub Actions template in references/ci-github-actions.md includes hardcoded credentials (postgres:test) for a local PostgreSQL service used in end-to-end testing. These are example credentials for an ephemeral test environment defined within the same workflow and do not represent a leak of actual secrets.

pflow-golang-setup