Crucible Studio

Crucible Studio is the VS Code operator surface for Enkari Crucible — adversarial audits, multi-host orchestration, and the full 436 skills · 78 agents · 64 numbered prompts · 38 industry packs methodology substrate, brought into the editor with status-bar context, tree views, chat launch points, diagnostics, scheduler controls, and release-evidence commands.

Claude-first: every routed turn runs on your own Claude Code by default. Operators who want cross-vendor adversarial passes can flip on multi-model to add OpenAI Codex/GPT — consistent project discipline either way, from one VS Code surface.

Pricing

Install free from the VS Code Marketplace, then start a free 30-day trial at enkari.com/crucible-studio: enter your email and an activation key arrives by return email. Studio is inert until you activate it.

After the trial, one subscription is $499/month or $4,990/year (~17% off annually — two months free). The subscription authorizes host dispatch; there is no Studio-side API-call metering. Enterprise contracts (white-label, SSO, custom industry packs) are by Statement of Work.

Studio keeps its logs, memories, REFLEXION rows, local telemetry, and activation file on your machine unless you configure an explicit sync/export destination. Full feature list, checkout status, and refund/discount terms are at enkari.com/crucible-studio.

Provider posture

Studio is model-agnostic at the methodology layer and Claude-first in practice: auto-routing runs Claude Code only unless multi-model is enabled, at which point audit/strategy turns may add an OpenAI Codex/GPT adversarial pass. Both run through their official CLIs under your own accounts. Cursor, ChatGPT, Gemini, and other host surfaces are treated as host-adapter guidance surfaces unless/until they expose a supported headless dispatch path.

Overview

The extension surfaces:

• 436 skills, 78 agents, 64 numbered prompts, and 38 industry packs
• Skills, agents, prompts, hooks, and industry-pack tree views
• Live context refresh and retrieval commands
• Bootstrap and runtime health checks
• Scheduler status, with autonomy disabled by default
• Audit, SBOM, bundle, and release-readiness command entry points
• Cold-load budget diagnostic — surface session-start context cost (plan-mode auto-inject + foundation prefix + project memory) with operator-actionable remediation recommendations

The extension is a companion to the @enkari/crucible package. The CLI owns substrate installation and verification; Studio gives operators a visual control plane inside VS Code.

Install

1. Install Enkari Crucible Studio from the VS Code Marketplace — click Install on the Marketplace page, or search "Enkari Crucible Studio" in the VS Code Extensions view.
2. Get an activation key: start a free 30-day trial at enkari.com/crucible-studio (enter your email; the key arrives by email), or start a paid subscription for a permanent key.
3. In VS Code, open the Command Palette and run Enkari Crucible: Paste activation key, then paste your key. No restart needed.
4. Open the Crucible view container and run Crucible: Refresh Live Context to confirm activation.

Activation is privacy-preserving and machine-bound. Studio verifies the key locally, then refreshes seat status with Enkari's activation service so one subscription seat stays active on one machine at a time. The check sends only activation and machine metadata.

When the operator explicitly dispatches work to Claude Code or OpenAI Codex/GPT, Studio assembles a bounded prompt for the configured local host CLI. Depending on settings and task, that prompt can include the user request, selected file excerpts, retrieval citations, redacted session history, foundation guidance, and selected Crucible prompt/skill/agent instructions. Studio does not upload project code to Enkari by default; model-provider data handling is governed by the host CLI/provider the operator configures.

Use the Crucible anywhere — not only the Studio chat

Activation wires the Crucible into your existing AI terminals, so the Studio chat panel is one surface among several:

• Claude Code terminals — after activation, type /crucible in any claude session to activate the Crucible there (add a task inline: /crucible audit this repo). The skill calls the server-side crucible_kickoff tool, which returns the session activation brief — no local Crucible files are needed. To (re)install the skill manually, run Crucible: Enable /crucible in Claude Code terminals from the Command Palette.
• Any MCP-capable host — activation registers the enkari MCP server (mcp.enkari.com) in ~/.claude.json (and ~/.codex/config.toml when enabled). Hosts connected to it can call crucible_kickoff, crucible_skill_retrieve, the adversarial chain tools, the factoid register, and trust receipts directly.
• VS Code native chat — mention @crucible in the built-in chat view.

If you work without the local substrate (the normal packaged install), crucible_kickoff is the launch point — it replaces the substrate-side KICKOFF.md ritual with one tool call.

Developer / substrate-side install

For operators running the full substrate locally (bootstrap, audit scripts, custom packs):

1. Install the @enkari/crucible package or clone the substrate.

2. Run:

   crucible bootstrap
   

3. Install the extension from the Marketplace, or load a locally-built .vsix for development.

4. Open the Crucible view container in VS Code.

5. Run Crucible: Refresh Live Context.

For local extension development:

npm ci
npm run prepublish-vsix
npm run ci:verify

Substrate path configuration (required for packaged installs)

After installing the packaged extension, Crucible Studio must locate the substrate root — the directory containing scripts/, agents/, skills/, and the rest of the @enkari/crucible package. The extension walks candidates in this order and uses the first one whose scripts/ subdir exists:

1. PROMPTS_DIR environment variable
2. CRUCIBLE_PROMPTS_DIR environment variable
3. VSCode workspace setting crucible.promptsDir
4. Extension-adjacent path (<extension-install>/../..) — only resolves to a valid substrate in dev mode where the extension is symlinked into <substrate>/packages/crucible-studio/

For packaged installs, candidate 4 resolves to ~/.vscode/, which has no scripts/ subdir. Without configuring 1, 2, or 3, Studio commands that invoke substrate scripts (drift report, twin retire / restore / embed / drift, factoid add / stats / contest / retire, trust receipt emit, weekly hallucination regression, render face sheet, render twin profile, voice harvester, etc.) will surface an actionable error toast and decline to run.

Recommended configuration

Pick one of:

• VSCode user settings (per-machine, survives extension reinstall):

  {
    "crucible.promptsDir": "C:\\path\\to\\your\\Crucible\\substrate"
  }
  

• Environment variable (per-shell or per-user):

  # Windows PowerShell (persistent, current user)
  setx CRUCIBLE_PROMPTS_DIR "C:\path\to\your\Crucible\substrate"
  

  # macOS / Linux (add to ~/.zshrc or ~/.bashrc)
  export CRUCIBLE_PROMPTS_DIR="$HOME/path/to/your/Crucible/substrate"
  

Restart VSCode after setting either.

Quick verification (no Studio launch required) — from the substrate root:

npm run diagnose:models

Reports scanned=73 translated=N unchanged=M. On a healthy install where the installed agents match the host table, translated=0 unchanged=73. If the output is non-zero translated, the agent installer needs to run (npm run install:agent-models).

In-Studio verification: restart VSCode, open the Command Palette, run Crucible: Render face sheet — it should open a terminal and execute <substrate>/scripts/face-sheet-render.mjs.

If you see Crucible: substrate scripts/ directory not found, the resolver walked all four candidates and none had a scripts/ subdir. Double-check the configured path actually points at the substrate root, not a parent or child.

Autonomy

Scheduled autonomous evidence tasks are opt-in. They do not run by default.

To opt in, enable the Studio setting crucible.autonomy.enabled or set:

CRUCIBLE_AUTONOMY=on

When autonomy is disabled, scheduler commands can still report status and explicit operator-run commands can still execute.

Cold-load discipline

Long Crucible sessions on Claude Code or OpenAI Codex/GPT can consume 600k–900k tokens of session-start context if plan files, foundation prefix, and project memory grow unchecked. The current Studio ship includes the Crucible: Diagnose cold-load budget command (Command Palette) which spawns the substrate's unified diagnostic against the active workspace folder.

Studio Chat also applies a Static Payload + Dynamic Compaction policy before host dispatch:

• stable foundation files, project memory, pack context, and retrieved docs stay in the prompt prefix so provider caches can hit
• active conversation history is treated as dynamic working memory
• crucible.chat.autoCompactAfterTokens defaults to 150000 estimated dynamic-thread tokens
• crucible.chat.contextWindowTokens defaults to 1000000; crucible.chat.dynamicContextWindowShare lets operators tune the pre-send compaction threshold for different hosts
• when the dynamic thread crosses the effective threshold (the lower of token budget and configured window share), Studio archives older events under .crucible/sessions/.compaction, carries forward a scoped summary, clears host resume ids, and sends the current turn against the compacted Studio context
• the active session strip shows the estimated dynamic context budget so operators can compact before the host window gets noisy
• the carried summary uses a dense state-packet shape and redacts local paths/secrets before it can be injected back into host context

Substrate-side autonomous-fired sentinels surface regrowth without operator action:

• weekly-plan-file-size-sentinel — warns when any ~/.claude/plans/*.md exceeds 500 KB
• daily-session-context-budget-gate — surfaces session-start budget per the canonical 5-band model
• weekly-per-project-context-budget-audit — measures cold-load across all registered projects

Full discipline guide: docs/COLD_LOAD_DISCIPLINE.md in the substrate repository.

License

Crucible Studio is proprietary commercial software. Use is governed by the Enkari Crucible license and EULA. See the repository-level LICENSE.md and EULA.md.

Support

• Sales / subscriptions: sales@enkari.com
• Support: support@enkari.com
• Security: security@enkari.com
• Legal: legal@enkari.com