The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes shell scripts (scripts/tickets.sh and scripts/map.sh) to perform its core functions. These scripts execute the GitHub CLI (gh) and a bundled Python adapter. The Python code is executed via python3 from a heredoc, which is a standard method for integrating complex logic into shell-based skills.
[INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it ingests and processes content from external issue trackers that could be controlled by third parties.
Ingestion points: Untrusted content such as issue titles, bodies, and comments are retrieved using scripts/tickets.sh and scripts/map.sh across all supported trackers (GitHub, Linear, Jira).
Boundary markers: There are no explicit boundary markers or instructions to the model to ignore embedded commands within the ingested data during the reconciliation (references/reconcile.md) or reporting (references/report.md) phases.
Capability inventory: The skill possesses significant capabilities, including the ability to write to the local file system (roadmap updates), create new tracker issues, and modify existing issue metadata (labels, dependencies, and comments) via its bundled scripts.
Sanitization: The skill performs basic sanitization of inputs for its internal tools (e.g., JQ escaping in scripts/tickets.sh and ADF formatting for Jira), but it does not filter or sanitize natural language content for potential prompt injection attacks.

vision