The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes multiple local shell scripts located in the user's home directory under ~/.claude-octopus/plugin/scripts/. These scripts, including orchestrate.sh, octo-state.sh, state-manager.sh, and check-providers.sh, manage the workflow state and interface with external AI providers for code and document validation.
[DATA_EXFILTRATION]: The skill interacts with the GitHub CLI (gh) and a local wrapper script safe-gh-comment.sh to post validation report summaries as comments to GitHub pull requests. This involves transmitting project-related metadata and validation findings to GitHub's official platform.
[INDIRECT_PROMPT_INJECTION]: The skill ingests context from previous project phases and local state files to inform the validation process, which presents a surface for indirect prompt injection.
Ingestion points: Project state is retrieved from .octo/STATE.md and prior phase context (discovery, definition, development) is read via the state-manager.sh script.
Boundary markers: User prompts and validation requests are passed as arguments to the shell execution tool.
Capability inventory: The skill maintains the ability to execute shell scripts and perform network operations through the GitHub CLI.
Sanitization: The skill instructions do not explicitly describe sanitization or validation of the project context data before it is processed by the orchestration scripts and external models.

flow-deliver