The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill implements a workflow where the main agent reads reports generated by subagents and interpolates their findings into new task prompts. This creates a surface where a subagent could potentially inject instructions into a report that are later executed by another agent in the chain.
Ingestion points: The main agent reads .claude/reports/_registry.md and specific report files within the .claude/reports/ directory as described in the 4-Step Workflow (SKILL.md).
Boundary markers: The Task Prompt Template provided in templates.md uses structural headers like **Context from registry:** and **Requirements:**. While these provide structure, they lack strong, unique delimiters or explicit instructions to the subagent to ignore instructions contained within the interpolated context.
Capability inventory: The skill is authorized to use Bash, Write, and Edit tools. It includes scripts for file verification (verify.sh) and report archiving (archive_reports.py).
Sanitization: The instructions do not define any mechanism for sanitizing, escaping, or filtering content from previous reports before it is interpolated into the context of a new task.
[COMMAND_EXECUTION]: The skill utilizes the Bash tool to execute internal scripts for system maintenance and verification.
Evidence: The workflow (SKILL.md) and reference documentation (reference.md) explicitly instruct the agent to run .claude/skills/agent-coordination/scripts/verify.sh and python .claude/skills/agent-coordination/scripts/archive_reports.py. These scripts perform file system operations such as moving files, creating directories, and checking file content length.

agent-coordination