ArgosBrain ships exactly two services. Card 1 runs invisibly, every edit, via hooks. Card 2 runs on-demand, weekly or pre-merge, via a single slash command. Same engine — sub-millisecond P99, $0 per query, local-first — two completely different surfaces.
Your agent stops hallucinating APIs, leaves no stubs, breaks no callers. Automatic in Claude Code (PreToolUse + PostToolUse hooks fire deterministic, the agent can't refuse). Prompt-driven in Cursor / Aider / Cline (auto-installed rules — the agent is strongly guided, compliance ~90-95% on modern models).
mcp__argos__preflight with the symbol the agent is about to touch. Returns existence + total caller count + production/test split + risk verdict (LOW/MEDIUM/HIGH). Injected into the agent's context before the Edit fires.todo!(), Python raise NotImplementedError, JS throw new Error("not implemented"), etc). Hit → agent obligated to fix before claiming "done".user.send_email() when the actual method is user.email() — hallucinated API.todo!() in 3 of 7 branches.defaults.ts and breaking 2 test files that consume it through anonymous arrow callbacks.throw new Error("not implemented") or unimplemented!() then declaring "Done!".| Agent runtime | Enforcement | Compliance | What ships at argosbrain init |
|---|---|---|---|
| Claude Code | Deterministic | 100% | PreToolUse + PostToolUse hooks. The agent never sees the choice — hooks fire before / after every Edit regardless. |
| Cursor | Prompt-driven | ~90-95% | .cursor/rules/argosbrain.mdc with alwaysApply: true. Injected in every chat; Opus 4.7 follows ~95%. |
| Aider / Cline / Codex CLI | Prompt-driven | ~80-90% | CONVENTIONS.md · .clinerules · ~/.codex/prompts/. Compliance depends on the model. |
| Raw MCP host (custom) | MCP instructions only | varies | MCP server's instructions field — universal signal every MCP-compatible agent reads. |
For deterministic enforcement, we recommend Claude Code. Cursor / Aider / Cline still ship the value via auto-applied rules; just not 100%.
cd ~/your-project
argosbrain init --install-config .
# Drops 5 hooks in ~/.claude/hooks/, registers them per-project,
# writes .cursor/rules/argosbrain.mdc + .clinerules + CONVENTIONS.md
# + CLAUDE.md. Restart your IDE. Done.Nine attacker perspectives. Static control-flow only — PoC pseudocode for your team's verification, no live exploits. Findings ranked by (impact × confidence) / cost-to-exploit and composed into MITRE ATT&CK + Unified Kill Chain + OWASP-mapped narratives. One slash command. 30-50 minutes. Run weekly or before every major merge.
Plus Perspective 10 — Chain Composer — turning the nine perspectives' flat primitives into ranked kill chains, mapped to Cyber Kill Chain + MITRE ATT&CK + Unified Kill Chain phases, and tied to historical campaigns.
# In Claude Code / Cursor / Aider / Cline, type:
/argos-security
# 9 perspectives run in parallel. 30-50 min on a 50k LOC codebase.
# Output: ranked kill chains, MITRE ATT&CK mapping, PoC pseudocode.Static control-flow review only. PoC pseudocode for your team's verification — no live exploits. IaC coverage is structurally limited; pair with Prowler / Checkov / Pacu for deeper cloud reasoning.
ArgosBrain ships a code verification engine — call graphs, reachability, ~59 MCP tools, 44 skills under the hood. We could sell every primitive as a service (and used to: 14 services, each its own pitch). We learned that's the wrong shape: agents drown in tool catalogs, buyers can't pick, the whole offering reads as broad-but-shallow.
So we ship two services that compose every primitive internally. The deeper toolkit is available to power users (argosbrain init --skill-pack full, ARGOSBRAIN_EXPOSE=full) — but you don't have to know any of that to use the two services above.
If your fire isn't covered by these two, tell us. Compliance audit prep, M&A code DD, refactoring blast-radius, smart-contract reachability — we have the primitives, we just don't sell them as separate cards. Email contact@argosbrain.com with the use case and we'll tell you which engine subset solves it, or surface it as a third card if demand justifies.
Each link below opens an opinionated install + tooling guide written for that role. Same Card 1 + Card 2 — different lead-in.
Sign in with GitHub → · How it works · Read the three papers · Skills detail · Tell us what's missing