The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits an Indirect Prompt Injection surface. It reads external content from GitLab issues and local markdown files (BC_SPEC.md, AGGREGATE_*.md) and interpolates this potentially untrusted data directly into a prompt template (feature-spec-issue-template.md).
Ingestion points: Reads GitLab issue content via mcp__gitlab__get_issue and local domain documentation via file read operations.
Boundary markers: None observed. The content is directly mapped to placeholders like {{feature_summary}} and {{story_title}} without delimiters or instructions to ignore embedded commands.
Capability inventory: The skill has the capability to write to the external GitLab environment using mcp__gitlab__create_issue and mcp__gitlab__create_issue_link.
Sanitization: No explicit sanitization or validation of the input data is mentioned before it is used to generate the new GitLab issue.
[COMMAND_EXECUTION]: The skill uses local shell-like commands to read project files. However, these operations are restricted to specific project paths (docs/contexts/) and appear to be part of the intended workflow without exposure to arbitrary user-supplied arguments beyond a context slug.

feature-specification