heating the forge...

@forgecat/msitarzewski_agency-agents_agency-project-management — Security Risk Report | ForgeCat

msitarzewski_agency-agents_agency-project-management

Security risk report for @forgecat/msitarzewski_agency-agents_agency-project-management v0.0.11

Low

Profile sourced from public GitHub repository (msitarzewski/agency-agents) with MIT license.
Original commit hash and repository URL provided for transparency and auditability.

Low

All agent instructions focus on legitimate project management workflows: experiment design, Jira-linked Git discipline, cross-functional coordination, operations optimization, portfolio management, and task specification.
No instructions to read credentials, exfiltrate data, reveal system prompts, or execute remote payloads.
No guidance poisoning: agents recommend standard practices (statistical rigor, Git hygiene, stakeholder communication, process documentation) without deceptive or security-weakening directives.

Details

Findings

Invisible characters detected
agents/project-management-jira-workflow-steward.md · SKILL-011
Invisible characters detected
agents/project-management-project-shepherd.md · SKILL-011

Evidence

Experiment Tracker: 'Always calculate proper sample sizes before experiment launch' and 'Ensure random assignment and avoid sampling bias' — standard scientific methodology.
Jira Workflow Steward: 'Never place secrets, credentials, tokens, or customer data in branch names, commit messages, PR titles, or PR descriptions' — security-conscious guidance.
Project Shepherd: 'Provide honest, transparent reporting even when delivering difficult news' — ethical leadership, not manipulation.
Studio Operations: 'Document all processes with clear, step-by-step procedures' and 'Maintain version control for process documentation' — operational best practices.
Senior Project Manager: 'Don't add luxury or premium requirements unless explicitly in spec' and 'No background processes in any commands - NEVER append &' — realistic, safe implementation guidance.

Low

No tool definitions or MCP servers declared in profile; agents are pure instruction/personality profiles without executable permissions.
Agent instructions describe advisory and coordination roles (project planning, workflow governance, stakeholder communication) that do not require shell, file write/delete, or network access.
Jira Workflow Steward includes a bash validation hook as a TEMPLATE/EXAMPLE for repository maintainers to adopt — not a tool the agent itself executes; hook is read-only validation logic.

Details

Evidence

Profile declares 'Dependencies: None' and no MCP servers or tool configs.
All agents operate as conversational advisors: Experiment Tracker advises on statistical design, Jira Steward advises on Git discipline, Project Shepherd advises on coordination — none request execution authority.
Bash hook in Jira Steward is presented as 'Commit and Branch Validation Hook' — a reference implementation for teams to install in their own repositories, not a command the agent runs.

Low

No MCP servers defined in profile.
No external binary execution, network access, or filesystem permissions declared.
Profile is pure markdown instruction set with no server configuration.

Details

Evidence

Profile metadata lists 'Dependencies: None' and no MCP server definitions appear in any agent file.