The Agent Skills Directory

[COMMAND_EXECUTION]: Uses the official GitHub CLI (gh) to perform repository management tasks. While these commands are powerful, the skill explicitly defines restricted permissions (e.g., no closing issues, no modifying user content, no touching code). It requires the user to have already authenticated the gh CLI with appropriate scopes.
[PROMPT_INJECTION]: The skill is vulnerable to Indirect Prompt Injection (Category 8). It reads untrusted issue bodies and comments to perform triage and "rubric" evaluation.
Ingestion points: Reads number,title,labels,body,comments from GitHub issues via gh issue list (SKILL.md).
Boundary markers: None identified for distinguishing user content from instructions during processing.
Capability inventory: Can modify issue labels, update project board status, comment on issues, and append content to issue bodies using gh CLI.
Sanitization: No evidence of input sanitization or filtering for the data retrieved from issues.
[EXTERNAL_DOWNLOADS]: No external network operations or script downloads are initiated. All actions are performed via the local gh CLI against the configured GitHub repository.
[DATA_EXFILTRATION]: No evidence of data exfiltration. The skill interacts solely with the specified repository and project board as defined in the local configuration file (.claude/backlog-manager.yaml).
[SAFE]: The skill implements a robust "dry-run" mechanism by default, requiring explicit user input (apply) for any write operations, which significantly mitigates the impact of unexpected behaviors or injection attempts.

backlog-manager