The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's workflow orchestration logic depends on parsing external, potentially untrusted files to determine execution flow. ● Ingestion points: The skill reads milestone status from docs/{scope}-prd-task-plan.md and hook definitions from skills/auto-trigger/SKILL.md. ● Boundary markers: Absent. The skill uses grep and cat to extract data from these files without explicit delimiting or instruction-ignore markers. ● Capability inventory: The skill has access to Bash, Read, Write, Edit, and AskUserQuestion as defined in the skill frontmatter. ● Sanitization: Absent. There is no evidence of validation or sanitization of the trigger, mode, or context values read from the YAML metadata before they are used to drive agent behavior.
[COMMAND_EXECUTION]: The skill uses shell commands to detect workflow milestones and read configuration. The Implementation Steps section describes using grep, ls, and cat to check file statuses and read hook metadata. The interpolation of {scope} in file paths represents a potential path traversal surface if the variable is not strictly validated.

workflow-orchestrator