The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests and analyzes various project files created by users (such as GDDs, ADRs, and sprint plans) to determine if a project is ready to advance stages. This creates a surface where malicious content within these documents could attempt to influence the agent's logic.
Ingestion points: The skill reads from design/gdd/, docs/architecture/, production/sprints/, production/epics/, and production/qa/ as documented in SKILL.md and CONTRACT.md.
Boundary markers: The skill incorporates a 'Collaborative Protocol' requiring user confirmation before significant actions and report generation, acting as a human-in-the-loop check.
Capability inventory: The skill possesses capabilities for command execution (via Bash), file modification (updating project.yaml), and spawning sub-agents for specialized reviews.
Sanitization: Instructions include checks for empty headers and template placeholders, though it relies on manual verification for content adequacy.
[DYNAMIC_CONTEXT_INJECTION]: SKILL.md uses the !command syntax to execute a shell command during the skill's initialization to resolve project settings.
Evidence: !bash "${CLAUDE_SKILL_DIR}/../../hooks/yaml-helper.sh" resolve_config ... at the top of SKILL.md.
Context: This is a platform-specific feature used here for benign configuration resolution from local project scripts.
[COMMAND_EXECUTION]: The skill relies on several local bash scripts to perform automated project audits and state updates.
Evidence: Execution of .claude/scripts/artifact-check.sh, .claude/scripts/gdd-structure-check.sh, .claude/scripts/adr-dep-graph.sh, and .claude/scripts/set-stage.sh.
Scope: These operations are restricted to scripts located within the project's own .claude/ and hooks/ directories.

gate-check