The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to Indirect Prompt Injection (Category 8). It ingests untrusted data from external sources including Linear issue comments, PR descriptions, and git commit messages. These sources could be used by an adversary to embed malicious instructions aimed at influencing the reconciliation process.
Ingestion points: Linear comments, issue descriptions, PR bodies, git commit messages, and local agent session logs.
Boundary markers: The skill defines a structured YAML schema for its reconciliation comments to help distinguish its output from untrusted content, though it does not explicitly outline input sanitization logic.
Capability inventory: The skill uses shell tools to perform git deletions and PR updates, and uses the Linear API to modify issue statuses and comments.
Sanitization: The instructions mitigate risks by enforcing a 'do not trust any single source' rule and requiring explicit human confirmation before any destructive or history-changing actions (e.g., deleting worktrees, force-pushing).
[COMMAND_EXECUTION]: The skill uses shell commands via git and gh to manage the repository state. This includes executing deletions and rebases based on discovered evidence.
Evidence: The skill instructions provide detailed logic for determining 'main equivalence' using ancestry and patch-id checks before recommending branch deletion, ensuring operations are performed on validated targets.

linear-repo-reconcile