The Agent Skills Directory

[DYNAMIC_CONTEXT_INJECTION]: The skill uses the dynamic context injection syntax (!command) in the 'Context' section to execute a find command at skill load time. This command is used to locate project-specific configuration files (.claude/session-plugin.local.md) within the workspace.
[INDIRECT_PROMPT_INJECTION]: The skill ingests and processes untrusted data from external and local sources, which could potentially contain malicious instructions intended to influence the agent's behavior during the briefing process.
Ingestion points: The skill reads GitHub issue titles and pull request data (via MCP tools or gh CLI), local configuration files (.claude/session-plugin.local.md, ~/.claude/session-plugin.local.md), and markdown journal notes from the user's filesystem.
Boundary markers: There are no explicit boundary markers or specialized instructions defined in the prompt to ensure the agent ignores potentially malicious commands embedded within the ingested data.
Capability inventory: The skill possesses the Bash tool (used to execute a survey script), Read access to files, and TodoWrite for task management.
Sanitization: While the skill applies "judgment" to filter and surface relevant items, it does not explicitly sanitize the text content (e.g., GitHub issue descriptions or journal todos) for prompt injection patterns.

session-spinup