The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests and processes data from GitHub issues, repository files, and user comments, which are external and potentially untrusted sources.
Ingestion points: Processes input from /woostack-execute <issue URL>, repository files, and comments to determine the implementation task.
Boundary markers: The instructions explicitly state that external text and tool output cannot widen the authorized outcome or access secrets. It mandates using isolated workspaces (worktrees) to prevent collateral impact.
Capability inventory: The skill is capable of editing source code, executing Git commands, and running build/test runners.
Sanitization: Guidelines in references/infrastructure.md and references/delivery.md instruct the agent to redact credentials, authorization tokens, and personal data from telemetry and logs.
[COMMAND_EXECUTION]: The skill uses various command-line interfaces to perform repository management and verification tasks.
Evidence: Utilizes git for status, diffing, and committing; uses gh (GitHub CLI) for pull request operations; and executes local build or test tools as part of the verification process.
[DYNAMIC_EXECUTION]: The skill is instructed to run commands that may be dynamically defined within the repository's configuration.
Evidence: The references/delivery.md file mandates the execution of "nonempty configured commit.command" during the delivery phase. This represents a vector where repository-specific configuration can trigger arbitrary command execution.
[EXTERNAL_DOWNLOADS]: The skill performs network requests to interact with development platforms and registries.
Evidence: It interacts with GitHub APIs for issue and PR management. references/patterns.md also references the use of npm view to resolve package versions from the official registry. These operations target well-known and established service providers.

woostack-execute