The Agent Skills Directory

[PROMPT_INJECTION]: The skill contains an indirect prompt injection surface by ingesting data from external sources (Jira and Confluence) which may contain untrusted content.
Ingestion points: Jira issue summaries, descriptions, acceptance criteria, and Confluence page content (SKILL.md).
Boundary markers: Absent; external content is treated directly as task context without delimitation.
Capability inventory: The agent can execute local git and tree commands based on the context received.
Sanitization: No explicit sanitization or validation of the retrieved external content is performed before processing.
[COMMAND_EXECUTION]: The skill instructs the agent to execute local system commands to gather repository information.
Evidence: tree -L 3 -I 'node_modules|__pycache__|.git|dist|build' and Atlassian MCP tool calls (mcp__atlassian__getJiraIssue, mcp__atlassian__getConfluencePage).
[DYNAMIC_CONTEXT_INJECTION]: The skill utilizes command substitution at load time to populate the agent's context with repository metadata.
Evidence: !git ls-files, !git log -10 --oneline, and !git status in SKILL.md.
Analysis: These are standard development operations used for orientation and do not involve sensitive data exfiltration or unsanitized user input in the command strings.

prime-codebase