heating the forge...

@forgecat/obra_superpowers — Security Risk Report | ForgeCat

obra_superpowers

Security risk report for @forgecat/obra_superpowers v0.0.12

Low

Static scanner assigned low-level source risk; profile is published from a known GitHub repository (https://github.com/obra/superpowers) with clear authorship and version tracking.
Metadata includes original commit hash (b36e0829c6d0140e93cfef2ca599b1b07d4a7797), MIT license, and multi-platform compatibility documentation, indicating transparent provenance.
No evidence of typosquatted dependencies, hidden binary execution, or supply-chain manipulation in the declared content.

Low

Profile content describes a software development methodology and skill library; it documents workflows (brainstorming, TDD, code review, debugging) without instructing the agent to ignore its instructions, exfiltrate credentials, or reveal system prompts.
Skills are framed as optional tools the agent may invoke (e.g., 'Use when facing 2+ independent tasks'), not as mandatory overrides of the agent's core behavior or hidden directives.
No evidence of guidance poisoning: the profile does not plant malicious knowledge, recommend typosquatted packages, embed backdoored code templates, or instruct weakening of security defaults. Recommendations for TDD, code review, and systematic debugging are legitimate software engineering practices.

Details

Findings

Invisible characters detected
skills/subagent-driven-development/SKILL.md · SKILL-011
Invisible characters detected
skills/subagent-driven-development/task-reviewer-prompt.md · SKILL-011

Evidence

brainstorming/SKILL.md: 'Do NOT invoke any implementation skill, write any code, scaffold any project, or take any implementation action until you have told your human partner what you intend and they have approved it.' — enforces approval gates, not manipulation.
receiving-code-review/SKILL.md: 'Verify before implementing. Ask before assuming. Technical correctness over social comfort.' — instructs critical evaluation, not blind obedience.
No instructions to read ~/.ssh, ~/.aws, .env, or exfiltrate environment variables; no instructions to hide or deny the profile's own instructions; no instructions to install remote payloads.

PRM-003Medium

hook-no-tool-match in ./hooks/generated-cursor-sessionstart-1.sh
Profile declares no tools or MCP servers; all skills are markdown-based guidance documents that instruct the agent to use existing tools (git, test runners, shell) in structured workflows.
Skills reference subagent dispatch and code review, which are medium-risk categories (subagent), but the profile does not grant new permissions—it documents how to use tools the agent already has access to.
Authority is scoped to software development workflows (planning, testing, debugging, code review) and matches the stated purpose of a 'superpowers software development methodology.'

Details

Findings

Evidence

hook-no-tool-match @ ./hooks/generated-cursor-sessionstart-1.sh
Profile metadata: 'Dependencies: None required to install this profile. Individual skills may require tools or credentials documented in their own SKILL.md files.'
Skills invoke existing agent capabilities (git, test runners, subagent dispatch) without requesting new high-risk categories (shell_write, file_delete, unrestricted network).

Low

Profile declares no MCP servers; all content is markdown-based skill documentation and prompt templates.
No hidden instructions in tool descriptions, no arbitrary binary execution, no unrestricted network or filesystem access defined.
MCP risk is not applicable to this profile.

Details

Evidence

Profile metadata: 'MCP servers (untrusted data): (none)'
All skills are text-based guidance; no server definitions, no tool bindings, no external service integrations declared.