The Agent Skills Directory

[COMMAND_EXECUTION] (MEDIUM): The skill executes multiple local shell scripts (discovery.sh, write-session-state.sh, system-check.sh). While these appear to be internal workflow tools, they perform file system discovery and state management.
[PROMPT_INJECTION] (LOW): The SKILL.md file contains instructions like 'CRITICAL: This guidance is mandatory' and 'ZERO OUTPUT RULE'. These are standard structural instructions for agent personas but match patterns used to constrain or override default behavior.
[DATA_EXFILTRATION] (SAFE): No evidence of network operations or exfiltration of sensitive files was detected. Network operations are not present in the provided scripts.
[EXTERNAL_DOWNLOADS] (SAFE): No remote packages or external scripts are downloaded; all referenced scripts are local to the project directory.
[INDIRECT_PROMPT_INJECTION] (LOW): The skill processes external Markdown and YAML files (specification.md, plan.md) to drive its logic.
Ingestion points: Reads frontmatter and content from docs/workflow/specification/ and docs/workflow/planning/ via discovery.sh.
Boundary markers: None identified in the script or instructions to delimit untrusted file content from system instructions.
Capability inventory: Uses Bash to execute local scripts and saves session state via write-session-state.sh.
Sanitization: None identified; the skill parses status, type, and format fields directly from file frontmatter and uses them for routing logic.

start-planning