The Agent Skills Directory

[SAFE]: The skill defines a mechanical workflow for project management using local file storage in the .plans/ and _docs/ directories. These operations are restricted to the project workspace and do not involve unauthorized file access or data exfiltration.
[SAFE]: The "Cast" section allows for structured multi-agent collaboration by assigning roles like Orchestrator and Reviewer. This is a workflow management feature and does not instruct the agent to bypass security filters or safety protocols.
[SAFE]: All external tool and model references, such as rjv-ollama-delegate and rjv-gated-build, are associated with the author's own ecosystem and do not involve untrusted third-party dependencies.
[PROMPT_INJECTION]: The skill contains an indirect prompt injection surface as it instructs the agent to read and follow instructions from local plan files which could be modified by external processes.
Ingestion points: The agent reads .plans/RESUME.md, .plans/<name>.md, and _docs/ files to determine its next actions.
Boundary markers: The skill uses literal markers such as >>> RESUME HERE <<< to locate the current task, but it lacks specific security delimiters to ignore potentially malicious instructions embedded in the project files.
Capability inventory: The agent is authorized to read/write files and verify project status against the code and database.
Sanitization: No sanitization of plan file content is performed. However, this risk is assessed as safe in the context of a development-focused workflow where the files are part of the repository being managed.
[NO_CODE]: This skill consists entirely of markdown instructions and does not bundle any executable scripts, binaries, or automated installation procedures.

rjv-feature-workflow