The Agent Skills Directory

[SAFE]: The skill facilitates a standard engineering planning process without introducing malicious patterns. It follows a controlled workflow and uses local files for context and output storage.
[DATA_EXFILTRATION]: No network operations or external data exfiltration patterns were identified in SKILL.md or the templates. The skill's operations are limited to reading local files and writing generated plans to specified project directories.
[COMMAND_EXECUTION]: The reference template (templates/story.md) includes commands for verification such as bun run lint, bun run typecheck, and bun test. These are standard development tools and are not executed from remote or untrusted sources.
[PROMPT_INJECTION]: The skill ingests untrusted data through user arguments and local project files, presenting a surface for indirect prompt injection.
Ingestion points: The $ARGUMENTS variable in SKILL.md and local project files (epics and source code) read during the planning process.
Boundary markers: Absent. External data is used directly within the agent context without delimiters.
Capability inventory: Writing markdown files to .agents/plans/ or planning/ directories and recommending the execution of local development commands.
Sanitization: None. The skill processes external content to generate a plan document.
Note: This ingestion surface is typical for planning tools and is considered safe in this context as the output is human-readable documentation and requires user review.

agile-story