The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and summarize untrusted data from the user's project environment to build context maps.
Ingestion points: Source code (controllers, services, and frontend components), database schema metadata via [DB_TOOL], OpenAPI specifications, git history, and Jira issue content via [ISSUE_TRACKER_TOOL].
Boundary markers: The templates provided for artifact generation (e.g., references/api.md) use structured headers but do not include explicit 'ignore embedded instructions' delimiters for the interpolated code content.
Capability inventory: The skill possesses capabilities to write to the local file system (.context/ directory, AGENTS.md), query the active database, and update issues within a Jira instance.
Sanitization: The skill mitigates risks by requiring the agent to document unverified facts in a ## Discovery Gaps section and mandating human approval for all updates, though it lacks technical sanitization of ingested strings.
[COMMAND_EXECUTION]: The skill invokes local shell commands to analyze the state of the repository.
Evidence: It executes git log --oneline -30 and git log --oneline -90 --stat to track recent changes and assess risk levels. These are standard read-only operations for gathering project context.

project-context