Collomia
Description
Collomia is a provider-neutral agentic coding agent for the terminal. Every model-proposed action passes through layered permissions, command analysis, OS sandboxing, network controls, an audit ledger, and evidence-based completion gates
Installation
This entry records only its repository, not the path inside it, so there is no
exact command to give. Open the source below and copy the folder into
~/.claude/skills/, or the file into ~/.claude/agents/.
README
Collomia

Collomia is a local-first, provider-neutral agent for the terminal. Its Developer profile combines a polished interactive interface with repository-aware tools and language-server intelligence; its Work profile supports research, analysis, automation, knowledge retrieval, document authoring, external actions, and direct questions in ordinary non-Git folders. Both use the same durable sessions, MCP and skills support, permission boundary, and headless JSONL interface. It ships as one `collo` binary for macOS, Linux, and Windows.
The project is built around an unusually explicit trust boundary. Model-proposed actions pass through layered permissions, command analysis, OS sandboxing, network controls, audit, and evidence-based completion checks. Standard execution is the permanent default. For work that benefits from a visible dependency graph, Orchestrated Goal adds optional evidence-gated durable execution in which the runtime—not model prose—owns readiness, evidence freshness, recovery, budgets, and the terminal outcome.
Why Collomia stands out
- Native containment on every major desktop OS. Commands can run under
macOS Seatbelt, Linux Landlock, or Windows AppContainer plus Job Objects.
Compatibility-first
automode reports degradation;requirefails closed when the requested boundary is unavailable. - Monotonic layered policy. Defaults, user configuration, trusted project configuration, and environment overrides compose predictably. A repository may tighten containment but cannot weaken the user's security posture, and project configuration, instructions, skills, and MCP servers remain quarantined until the workspace is trusted.
- Real network boundaries. Every sandbox backend can deny command network access. macOS also supports host-scoped sandbox egress through a Collomia-owned broker; Linux and Windows retain their honest all-or-nothing controls where
Related Skills
Fastapi Review
Review a FastAPI application for architecture, async correctness, dependency injection, Pydantic schemas, secu
Security Defense in Depth
Implement multi-layered testing and security best practices.
Security SecLists Official Repository
[OWASP Testing Guide](https://owasp.org/www-project-web-security-testing-guide/)
Security Threat Hunting with Sigma Rules
Use Sigma detection rules to hunt for threats and analyze security events
Security Maintenance Walkthrough - 2026-03-29
- Re-triaged the full 2026-03-15 security finding set against current `main` and wrote a fresh current-head re
Security Google Workspace Model Armor
Filter user-generated content for safety
Security Related Agents
Django Reviewer
Expert Django code reviewer specializing in ORM correctness, DRF patterns, migration safety, security misconfi
Token Auditor
Scans ui/src/ for hardcoded visual values, duplicate components, and shadcn replacement candidates; produces d
Gitnexus Security Boundary Reviewer
GitNexus security and trust-boundary reviewer. Use for auth, permissions, secrets, injection, unsafe parsing,