The Agent Skills Directory

[COMMAND_EXECUTION]: The skill invokes several local scripts (scripts/conflict-detect.sh, scripts/pr-coordinator.sh, scripts/dep-guard.sh) during the planning process. These scripts are used for conflict detection and dependency management, executing within the user's environment.
[DYNAMIC_EXECUTION]: The skill instructions define a pattern where the agent is expected to execute commands suggested by tool output, specifically identifying FORGE_AGENT_7f3a:PROMPT: directives in the output of forge team verify. This enables dynamic control flow based on external or tool-generated content.
[INDIRECT_PROMPT_INJECTION]: The skill ingests data from external sources including web research (via the research subskill) and the local codebase (commits, existing files). This data is processed by sub-agents to influence design and task generation. There are no explicit sanitization routines or boundary markers defined to isolate the agent context from potentially malicious instructions embedded in these inputs.
Ingestion points: Phase 1 Step 1 (codebase/commits), Phase 2 (web research via subskill), and forge sync (team state).
Capability inventory: Bash, Write, Edit, and forge CLI commands.
Boundary markers: Absent for ingested external content.
Sanitization: Absent for external or codebase inputs.

plan