OpenCode Swarm

OpenCode Swarm is a plugin for OpenCode that turns a single AI coding session into an architect-led team of 11 specialized agents. One agent writes the code. A different agent reviews it. Another writes and runs tests. Another checks security. Nothing ships until every required gate passes.

bunx opencode-swarm install

This single command installs the package, registers it as an OpenCode plugin, disables conflicting default agents, and creates a ready-to-edit config at ~/.config/opencode/opencode-swarm.json. Requires Bun (bun --version to check). If you must use npm: npm install -g opencode-swarm && opencode-swarm install.

Why Swarm?

Most AI coding tools let one model write code and ask that same model whether the code is good. That misses too much. Swarm separates planning, implementation, review, testing, and documentation into specialized internal roles — and enforces gated execution so agents never mutate the codebase in parallel.

Key Features

• 🏗️ 11 specialized agents — architect, coder, reviewer, test engineer, critic, critic_sounding_board, critic_drift_verifier, explorer, SME, docs, designer
• 🔒 Gated pipeline — code never ships without reviewer + test engineer approval
• 🔄 Phase completion gates — completion-verify and drift verifier gates enforced before phase completion
• 🔁 Resumable sessions — all state saved to .swarm/; pick up any project any day
• 🌐 20 languages — TypeScript, Python, Go, Rust, Java, Kotlin, C/C++, C#, Ruby, Swift, Dart, PHP, JavaScript, CSS, Bash, PowerShell, INI, Regex
• 🛡️ Built-in security — SAST, secrets scanning, dependency audit per task
• 🆓 Free tier — works with OpenCode Zen's free model roster
• ⚙️ Fully configurable — override any agent's model, disable agents, tune guardrails

You select a Swarm architect once in the OpenCode GUI. The architect coordinates all other agents automatically — you never manually switch between internal roles. If you use the default OpenCode Build / Plan modes, the plugin is bypassed entirely.

What Actually Happens

You say:

Build me a JWT auth system.

Swarm then:

1. Clarifies only what it cannot infer.
2. Scans the codebase to understand what already exists.
3. Consults domain experts when needed and caches the guidance.
4. Writes a phased implementation plan.
5. Sends that plan through a critic gate before coding starts.
6. Executes one task at a time through the QA pipeline:

• coder writes code
• automated checks run
• reviewer checks correctness
• test engineer writes and runs tests
• architect runs regression sweep
• failures loop back with structured feedback

1. After each phase, docs and retrospectives are updated.

All project state lives in .swarm/ — plans, evidence, context, knowledge, and telemetry. Resumable by design. If .swarm/ already exists, the architect goes straight into RESUME → EXECUTE instead of repeating discovery.

Execution Modes

Swarm has two independent mode systems:

Session modes — toggle per session with a slash command:

Project mode — persistent via execution_mode config key:

Switch session modes with /swarm turbo [on|off] or /swarm full-auto [on|off]. Set project mode in config. The two systems compose independently — see docs/modes.md.

Quick Start

→ For a complete first-run walkthrough, see Getting Started.

The 15-minute guide covers:

• Installation (bunx opencode-swarm install)
• Selecting the architect in OpenCode
• Running your first task
• Troubleshooting common issues

Commands

All 41 subcommands at a glance:

/swarm status              # Current phase and task
/swarm plan [N]            # Full plan or filtered by phase
/swarm agents              # Registered agents and models
/swarm diagnose            # Health check
/swarm evidence [task]     # Test and review results
/swarm reset --confirm     # Clear swarm state

See docs/commands.md for the full reference (41 commands).

The Agents

Swarm has 11 specialized agents. You don't manually switch between them — the architect coordinates automatically.

Run /swarm status and /swarm agents to see what's active.

How It Compares

LLM Provider Guide

Swarm works with any provider supported by OpenCode.

Free Tier (OpenCode Zen)

No API key required. Excellent starting point:

{
  "agents": {
    "coder": { "model": "opencode/minimax-m2.5-free" },
    "reviewer": { "model": "opencode/big-pickle" },
    "explorer": { "model": "opencode/trinity-large-preview-free" }
  }
}

Paid Providers

For production, mix providers by role:

Provider Formats

Model Fallback

Automatic fallback to a secondary model on transient errors:

{
  "agents": {
    "coder": {
      "model": "anthropic/claude-sonnet-4-20250514",
      "fallback_models": ["opencode/gpt-5-nano"]
    }
  }
}

See docs/configuration.md for full configuration reference.

Process Remediation Model (PRM)

Swarm monitors agent trajectories and injects course-correction guidance before loops form. Detects five failure patterns:

1. Repetition Loop — Same agent performs the same action repeatedly
2. Ping-Pong — Agents hand off back and forth without progress
3. Expansion Drift — Plan scope grows beyond original task
4. Stuck-on-Test — Coder and tests fail in a loop
5. Context Thrashing — Agent requests increasingly large file sets

When detected, escalation levels trigger:

• Level 1: Advisory guidance injected
• Level 2: Architect alert sent
• Level 3: Hard stop directive

Configure via:

{
  "prm": {
    "enabled": true,
    "pattern_thresholds": {
      "repetition_loop": 2,
      "ping_pong": 4,
      "expansion_drift": 3,
      "stuck_on_test": 3,
      "context_thrash": 5
    }
  }
}

Note: Some configuration fields (max_trajectory_lines, escalation_enabled) are defined in schema but not yet enforced at runtime.

Persistent Memory

.swarm/plan-ledger.jsonl — authoritative source of truth (v6.44 durability model)

.swarm/context.md — technical decisions and cached SME guidance

.swarm/evidence/ — review/test results per task

.swarm/telemetry.jsonl — session observability events (fire-and-forget, never blocks execution)

.swarm/curator-summary.json — phase-level intelligence and drift reports

Guardrails & Circuit Breakers

Every agent runs inside a circuit breaker that prevents runaway behavior:

Limits reset per task. Per-agent overrides available in config.

File Authority (Per-Agent Write Permissions)

Each agent can only write to specific paths:

• architect — Everything (except .swarm/plan.md, .swarm/plan.json)
• coder — src/, tests/, docs/, scripts/
• reviewer — .swarm/evidence/
• test_engineer — tests/, .swarm/evidence/
• explorer, sme — Read-only

Override via authority.rules in config.

Quality Gates

Built-in tools verify every task before it ships:

• syntax_check — Tree-sitter validation (12 languages)
• placeholder_scan — Catches TODOs, stubs, incomplete code
• sast_scan — 63+ security rules, 9 languages (offline)
• sbom_generate — Dependency tracking (CycloneDX)
• quality_budget — Complexity, duplication, test ratio limits

All tools run locally. No Docker, no network calls.

Context Budget Guard

Monitors how much context Swarm injects to prevent overflow:

• Warning threshold (70%) — Advisory when context reaches ~2800 tokens
• Critical threshold (90%) — Alert at ~3600 tokens with /swarm handoff recommendation
• Non-nagging — One-time alerts per session

Disable entirely with context_budget.enabled: false.

File Locking for Concurrent Safety

Hard lock on plan.json (serialized writes), advisory lock on events.jsonl (append-only log). Stale locks auto-expire via proper-lockfile.

Agent Categories

Agents are classified into four categories for the monitor server /metadata endpoint:

Use getAgentCategory(agentName) from src/config/agent-categories.ts to resolve an agent's category at runtime.

The Pipeline

Every task goes through this sequence. No exceptions, no overrides.

MODE: EXECUTE (per task)
│
├── 5a. @coder implements (ONE task only)
├── 5b. diff + imports (contract + dependency analysis + semantic diff context)
│       └── @system-enhancer injects AST-based semantic diff summary with blast radius
│           into @reviewer context (up to 10 files, conditional on declared scope)
├── 5c. syntax_check (parse validation)
├── 5d. placeholder_scan (catches TODOs, stubs, incomplete code)
├── 5e. lint fix → lint check
├── 5f. build_check (does it compile?)
├── 5g. pre_check_batch (4 parallel: lint, secretscan, SAST, quality budget)
├── 5h. @reviewer (correctness pass)
├── 5i. @reviewer (security pass, if security-sensitive files changed)
├── 5j. @test_engineer (verification tests + coverage ≥70%)
├── 5k. @test_engineer (adversarial tests)
├── 5l. architect regression sweep (scope:"graph" to find cross-task test regressions)
├── 5l-ter. test drift detection (conditional — fires when changes involve command behaviour,
│         parsing/routing logic, user-visible output, public contracts, assertion-heavy areas,
│         or helper lifecycle changes; validates tests still align with current behaviour)
├── 5m. ⛔ Pre-commit checklist (all 4 items required, no override)
└── 5n. Task marked complete, evidence written

If any step fails, the coder gets structured feedback and retries. After 5 failures on the same task, it escalates to you.

Architect Workflow Modes

The architect moves through these modes automatically:

CODEBASE REALITY CHECK (v6.29.2): Before any planning, the Architect dispatches Explorer to verify the current state of every referenced item. Produces a CODEBASE REALITY REPORT with statuses: NOT STARTED, PARTIALLY DONE, ALREADY COMPLETE, or ASSUMPTION INCORRECT. This prevents planning against stale assumptions. Skipped for greenfield projects with no existing codebase references.

Phase Completion Gates (v6.33.4): Before a phase can be marked complete, two mandatory gates are enforced: (1) completion-verify — deterministic check that plan task identifiers exist in source files, and (2) critic_drift_verifier evidence — verification that the drift verifier approved the implementation. Both gates are automatically bypassed when turbo mode is active.

Important

A second or later run does not necessarily look like a first run.

If .swarm/plan.md already exists, the architect may enter RESUME and then go directly into EXECUTE. That is expected and does not mean Swarm stopped using agents.

Use /swarm status if you are unsure what Swarm is doing.

Release automation uses release-please and requires conventional commit prefixes such as fix: or feat: on changes merged to main.

plan.md: Your Project Roadmap

# Project: Auth System
Current Phase: 2

## Phase 1: Foundation [COMPLETE]
- [x] Task 1.1: Create user model [SMALL]
- [x] Task 1.2: Add password hashing [SMALL]
- [x] Task 1.3: Database migrations [MEDIUM]

## Phase 2: Core Auth [IN PROGRESS]
- [x] Task 2.1: Login endpoint [MEDIUM]
- [ ] Task 2.2: JWT generation [MEDIUM] (depends: 2.1) ← CURRENT
  - Acceptance: Returns valid JWT with user claims, 15-minute expiry
  - Attempt 1: REJECTED — missing expiration claim
- [ ] Task 2.3: Token validation middleware [MEDIUM]

context.md: What's Been Decided

## Technical Decisions
- bcrypt cost factor: 12
- JWT TTL: 15 minutes; refresh TTL: 7 days

## SME Guidance (cached, never re-asked)
### security (Phase 1)
- Never log tokens or passwords
- Rate-limit login: 5 attempts / 15 min per IP

### api (Phase 1)
- Return 401 for invalid credentials (not 404)

Evidence Bundles

Every completed task writes structured evidence to .swarm/evidence/:

telemetry.jsonl: Session Observability

Swarm emits structured JSONL events to .swarm/telemetry.jsonl for observability tooling (dashboards, alerting, audit logs). Events are fire-and-forget — failures never affect execution.

{"timestamp":"2026-03-25T14:30:00.000Z","event":"session_started","sessionId":"abc123","agentName":"architect"}
{"timestamp":"2026-03-25T14:30:05.000Z","event":"delegation_begin","sessionId":"abc123","agentName":"coder","taskId":"1.1"}
{"timestamp":"2026-03-25T14:31:00.000Z","event":"delegation_end","sessionId":"abc123","agentName":"coder","taskId":"1.1","result":"success"}
{"timestamp":"2026-03-25T14:31:10.000Z","event":"gate_passed","sessionId":"abc123","gate":"reviewer","taskId":"1.1"}
{"timestamp":"2026-03-25T14:32:00.000Z","event":"phase_changed","sessionId":"abc123","oldPhase":1,"newPhase":2}

File rotates automatically at 10MB to .swarm/telemetry.jsonl.1.

The save_plan tool requires an explicit target workspace path. It does not fall back to process.cwd().

Explicit Workspace Requirement

• The working_directory parameter must be provided
• Providing no value or relying on implicit directory resolution will result in deterministic failure

Failure Conditions

Usage Contract

When using save_plan, always pass a valid working_directory:

save_plan({
  title: "My Project",
  swarm_id: "mega",
  phases: [{ id: 1, name: "Setup", tasks: [{ id: "1.1", description: "Initialize project" }] }],
  working_directory: "/path/to/project"  // Required - no fallback
})

Every agent runs inside a circuit breaker that kills runaway behavior before it burns your credits.

Limits reset per task. A coder working on Task 2.3 is not penalized for tool calls made during Task 2.2.

Architect Self-Coding Block

If the architect writes files directly instead of delegating to the coder, a hard block fires:

The counter resets only when a coder delegation is dispatched. This is a hard enforcement — not advisory.

Per-agent overrides:

{
  "guardrails": {
    "profiles": {
      "coder": { "max_tool_calls": 500, "max_duration_minutes": 60 },
      "explorer": { "max_tool_calls": 50 }
    }
  }
}

Swarm enforces per-agent file write authority — each agent can only write to specific paths. By default, these rules are hardcoded, but you can override them via config.

Default Rules

Prefixed Agents

Prefixed agents (e.g., paid_coder, mega_reviewer, local_architect) inherit defaults from their canonical base agent via stripKnownSwarmPrefix. The lookup order is:

1. Exact match for the prefixed name (if explicitly defined in user config)
2. Fall back to the canonical agent's defaults (e.g., paid_coder → coder)

{
  "authority": {
    "rules": {
      "coder": { "allowedPrefix": ["src/", "lib/"] },
      "paid_coder": { "allowedPrefix": ["vendor/", "plugins/"] }
    }
  }
}

In this example, paid_coder gets its own explicit rule, while other prefixed coders (e.g., mega_coder) fall back to coder.

Runtime Enforcement

Architect direct writes are enforced at runtime via toolBefore hook. This tracks writes to source code paths outside .swarm/ and protects .swarm/plan.md and .swarm/plan.json from direct modification.

Configuration

Override default rules in .opencode/opencode-swarm.json:

{
  "authority": {
    "enabled": true,
    "rules": {
      "coder": {
        "allowedPrefix": ["src/", "lib/", "scripts/"],
        "blockedPrefix": [".swarm/"],
        "blockedZones": ["generated"]
      },
      "explorer": {
        "readOnly": false,
        "allowedPrefix": ["notes/", "scratch/"]
      }
    }
  }
}

Rule Fields

Merge Behavior

• User rules override hardcoded defaults for the specified agent
• Scalar fields (readOnly) — user value replaces default
• Array fields (blockedPrefix, allowedPrefix, etc.) — user array replaces entirely (not merged)
• If a field is omitted in the user rule for a known agent (one with hardcoded defaults), the default value for that field is preserved
• If a field is omitted in the user rule for a custom agent (not in the defaults list), that field is undefined — there are no defaults to inherit
• allowedPrefix: [] explicitly denies all writes; omitting allowedPrefix entirely means no allowlist restriction is applied (all paths are evaluated against blocklist rules only)
• Setting enabled: false ignores all custom rules and uses hardcoded defaults

Custom Agents

Custom agents (not in the defaults list) start with no rules. Their write authority depends entirely on what you configure:

• Not in config at all — agent is denied with Unknown agent (no rule exists; this is not the same as "blocked from all writes")
• In config without allowedPrefix — no allowlist restriction applies; only any blockedPrefix, blockedZones, or readOnly rules you explicitly set will enforce limits
• In config with allowedPrefix: [] — all writes are denied

To safely restrict a custom agent, always set allowedPrefix explicitly:

{
  "authority": {
    "rules": {
      "my_custom_agent": {
        "allowedPrefix": ["plugins/", "extensions/"],
        "blockedZones": ["generated"]
      }
    }
  }
}

Advanced Examples

Glob Pattern Support

Use glob patterns for complex path matching:

{
  "authority": {
    "rules": {
      "coder": {
        "allowedGlobs": ["src/**/*.ts", "tests/**/*.test.ts"],
        "blockedGlobs": ["src/**/*.generated.ts", "**/*.d.ts"],
        "allowedExact": ["src/index.ts", "package.json"]
      },
      "docs_agent": {
        "allowedGlobs": ["docs/**/*.md", "*.md"],
        "blockedExact": [".swarm/plan.md"]
      }
    }
  }
}

Glob Pattern Features:

• ** — Match any number of directories: src/**/*.ts matches all TypeScript files in src/ and subdirectories
• * — Match any characters except path separators: *.md matches all Markdown files in current directory
• ? — Match single character: test?.js matches test1.js, testa.js
• Uses picomatch for cross-platform compatibility

Path Normalization and Symlinks:
Paths are resolved via realpathSync before matching, which resolves symlinks and prevents path-traversal escapes. However, if a symlink's target does not exist, realpathSync throws and the fallback returns the symlink's own path (unresolved). A dangling symlink inside an allowedPrefix directory will therefore pass prefix-based checks even if its intended target is outside the project. Use blockedExact or blockedGlobs to deny known dangling-symlink paths explicitly.

Evaluation Order:

1. readOnly check (if true, deny all writes)
2. blockedExact (exact path matches, highest priority)
3. blockedGlobs (glob pattern matches)
4. allowedExact (exact path matches, overrides prefix/glob restrictions)
5. allowedGlobs (glob pattern matches)
6. blockedPrefix (prefix matches)
7. allowedPrefix (prefix matches)
8. blockedZones (zone classification)

The Context Budget Guard monitors how much context Swarm is injecting into the conversation. It helps prevent context overflow before it becomes a problem.

Default Behavior

• Enabled automatically — No setup required. Swarm starts tracking context usage right away.
• What it measures — Only the context that Swarm injects (plan, context, evidence, retrospectives). It does not count your chat history or the model's responses.
• Warning threshold (0.7 ratio) — When swarm-injected context reaches ~2800 tokens (70% of 4000), the architect receives a one-time advisory warning. This is informational — execution continues normally.
• Critical threshold (0.9 ratio) — When context reaches ~3600 tokens (90% of 4000), the architect receives a critical alert with a recommendation to run /swarm handoff. This is also one-time only.
• Non-nagging — Alerts fire once per session, not repeatedly. You won't be pestered every turn.
• Who sees warnings — Only the architect receives these warnings. Other agents are unaware of the budget.

To disable entirely, set context_budget.enabled: false in your swarm config.

Configuration Reference

Example Configurations

Minimal (disable):

{
  "context_budget": {
    "enabled": false
  }
}

Default (reference):

{
  "context_budget": {
    "enabled": true,
    "max_injection_tokens": 4000,
    "warn_threshold": 0.7,
    "critical_threshold": 0.9,
    "enforce": true,
    "prune_target": 0.7,
    "preserve_last_n_turns": 4,
    "recent_window": 10,
    "tracked_agents": ["architect"],
    "enforce_on_agent_switch": true,
    "model_limits": { "default": 128000 },
    "tool_output_mask_threshold": 2000,
    "scoring": {
      "enabled": false,
      "max_candidates": 100,
      "weights": { "recency": 0.3, "relevance": 0.4, "importance": 0.3 },
      "decision_decay": { "mode": "exponential", "half_life_hours": 24 },
      "token_ratios": { "prose": 0.25, "code": 0.4, "json": 0.6, "logs": 0.1 }
    }
  }
}

Aggressive (for long-running sessions):

{
  "context_budget": {
    "enabled": true,
    "max_injection_tokens": 2000,
    "warn_threshold": 0.5,
    "critical_threshold": 0.75,
    "enforce": true,
    "prune_target": 0.6,
    "preserve_last_n_turns": 2,
    "recent_window": 5,
    "tracked_agents": ["architect"],
    "enforce_on_agent_switch": true,
    "model_limits": { "default": 128000 },
    "tool_output_mask_threshold": 1500,
    "scoring": {
      "enabled": true,
      "max_candidates": 50,
      "weights": { "recency": 0.5, "relevance": 0.3, "importance": 0.2 },
      "decision_decay": { "mode": "linear", "half_life_hours": 12 },
      "token_ratios": { "prose": 0.2, "code": 0.35, "json": 0.5, "logs": 0.05 }
    }
  }
}

What This Does NOT Do

• Does NOT prune chat history — Your conversation with the model is untouched
• Does NOT modify tool outputs — What tools return is unchanged
• Does NOT block execution — The guard is advisory only; it warns but never stops the pipeline
• Does NOT interact with compaction.auto — Separate feature with separate configuration
• Only measures swarm's injected context — Not the full context window, just what Swarm adds

Built-in Tools

All tools run locally. No Docker, no network calls, no external APIs.

Optional enhancement: Semgrep (if on PATH).

Gate Configuration

{
  "gates": {
    "syntax_check": { "enabled": true },
    "placeholder_scan": { "enabled": true },
    "sast_scan": { "enabled": true },
    "quality_budget": {
      "enabled": true,
      "max_complexity_delta": 5,
      "min_test_to_code_ratio": 0.3
    }
  }
}

Swarm uses file locking to protect shared state files from concurrent write corruption. The locking strategy differs by file: plan.json uses hard locking (write blocked on contention), while events.jsonl uses advisory locking (write proceeds with a warning on contention).

Locking Implementation

• Library: proper-lockfile with retries: 0 (fail-fast — no polling retries)
• Scope: Each tool acquires an exclusive lock on the target file before writing
• Agents: Lock is tagged with the current agent name and task context for diagnostics

Protected Files

Lock Semantics

The two protected tools use different strategies:

update_task_status — Hard lock on plan.json

When two calls contend for plan.json:

1. Exactly one call wins — only the first to acquire the lock proceeds
2. Winner writes — the lock holder writes to the file, then releases the lock
3. Losers receive success: false — with recovery_guidance: "retry" and an error message identifying the lock holder

{
  "success": false,
  "message": "Task status write blocked: plan.json is locked by architect (task: update-task-status-1.1-1234567890)",
  "errors": ["Concurrent plan write detected — retry after the current write completes"],
  "recovery_guidance": "Wait a moment and retry update_task_status. The lock will expire automatically if the holding agent fails."
}

What the caller should do: Retry update_task_status after a short delay.

phase_complete — Advisory lock on events.jsonl

When two calls contend for events.jsonl:

1. Lock is attempted — if acquired, write is serialized
2. If lock unavailable — a warning is added to the result and the write proceeds anyway
3. Both callers return success: true — duplicate concurrent appends are possible but events.jsonl is an append-only log and duplicate phase entries do not corrupt state

This asymmetry is intentional: plan.json stores mutable structured JSON where concurrent overwrites produce malformed files; events.jsonl is an append-only log where a duplicate entry is a recoverable nuisance.

Lock Recovery

If a lock-holding agent crashes or hangs, the lock file will eventually expire (handled by proper-lockfile stale-lock cleanup). On the next retry, the call will succeed. Swarm does not auto-retry on lock contention — the architect receives the error and decides when to retry.

Config file location: ~/.config/opencode/opencode-swarm.json (global) or .opencode/opencode-swarm.json (project). Project config merges over global.

{
  "agents": {
    "architect": { "model": "anthropic/claude-opus-4-6" },
    "coder": { "model": "minimax-coding-plan/MiniMax-M2.5", "fallback_models": ["minimax-coding-plan/MiniMax-M2.1"] },
    "explorer": { "model": "minimax-coding-plan/MiniMax-M2.1" },
    "sme": { "model": "kimi-for-coding/k2p5" },
    "critic": { "model": "zai-coding-plan/glm-5" },
    "reviewer": { "model": "zai-coding-plan/glm-5", "fallback_models": ["opencode/big-pickle"] },
    "test_engineer": { "model": "minimax-coding-plan/MiniMax-M2.5" },
    "docs": { "model": "zai-coding-plan/glm-4.7-flash" },
    "designer": { "model": "kimi-for-coding/k2p5" }
  },
  "guardrails": {
    "max_tool_calls": 200,
    "max_duration_minutes": 30,
    "profiles": {
      "coder": { "max_tool_calls": 500 }
    }
  },
  "authority": {
    "enabled": true,
    "rules": {
      "coder": {
        "allowedPrefix": ["src/", "lib/"],
        "blockedPrefix": [".swarm/"],
        "blockedZones": ["generated"]
      }
    }
  },
  "review_passes": {
    "always_security_review": false,
    "security_globs": ["**/*auth*", "**/*crypto*", "**/*session*"]
  },
  "automation": {
    "mode": "manual",
    "capabilities": {
      "plan_sync": true,
      "phase_preflight": false,
      "config_doctor_on_startup": false,
      "config_doctor_autofix": false,
      "evidence_auto_summaries": true,
      "decision_drift_detection": true
    }
  },
  "knowledge": {
    "enabled": true,
    "swarm_max_entries": 100,
    "hive_max_entries": 1000,
    "auto_promote_days": 30,
    "max_inject_count": 5,
    "dedup_threshold": 0.6,
    "scope_filter": ["global"],
    "hive_enabled": true,
    "rejected_max_entries": 200,
    "validation_enabled": true,
    "evergreen_confidence": 0.8,
    "evergreen_utility": 0.5,
    "low_utility_threshold": 0.2,
    "min_retrievals_for_utility": 3,
    "schema_version": "v6.17"
  }
}

Automation

Mode Detection (v6.13)

Swarm now explicitly distinguishes five architect modes:

• DISCOVER — After the explorer finishes scanning the codebase.
• PLAN — When the architect writes or updates the plan.
• EXECUTE — During task implementation (the normal pipeline).
• PHASE-WRAP — After all tasks in a phase are completed, before docs are updated.
• UNKNOWN — Fallback when the current state does not match any known mode.

Each mode determines which injection blocks are added to the LLM prompt (e.g., plan cursor is injected in PLAN, tool output truncation in EXECUTE, etc.).

Default mode: manual. No background automation — all actions require explicit slash commands.

Modes:

• manual — No background automation. All actions via slash commands (default).
• hybrid — Background automation for safe operations, manual for sensitive ones.
• auto — Full background automation.

Capability defaults:

• plan_sync: true — Background plan synchronization using fs.watch with debounced writes (300ms) and 2-second polling fallback
• phase_preflight: false — Phase preflight checks before agent execution (opt-in)
• config_doctor_on_startup: false — Validate configuration on startup
• config_doctor_autofix: false — Auto-fix for config doctor (opt-in, security-sensitive)
• evidence_auto_summaries: true — Automatic summaries for evidence bundles
• decision_drift_detection: true — Detect drift between planned and actual decisions

Plan Cursor (v6.13)

The plan_cursor config compresses the plan that is injected into the LLM context.

{
  "plan_cursor": {
    "enabled": true,
    "max_tokens": 1500,
    "lookahead_tasks": 2
  }
}

• enabled – When true (default) Swarm injects a compact plan cursor instead of the full plan.md.
• max_tokens – Upper bound on the number of tokens emitted for the cursor (default 1500). The cursor contains the current phase summary, the full current task, and up to lookahead_tasks upcoming tasks. Earlier phases are reduced to one‑line summaries.
• lookahead_tasks – Number of future tasks to include in full detail (default 2). Set to 0 to show only the current task.

Disabling ("enabled": false) falls back to the pre‑v6.13 behavior of injecting the entire plan text.

Tool Output Truncation (v6.13)

Control the size of tool outputs that are sent back to the LLM.

{
  "tool_output": {
    "truncation_enabled": true,
    "max_lines": 150,
    "per_tool": {
      "diff": 200,
      "symbols": 100
    }
  }
}

• truncation_enabled – Global switch (default true).
• max_lines – Default line limit for any tool output.
• per_tool – Overrides max_lines for specific tools. The diff and symbols tools are truncated by default because their outputs can be very large.

When truncation is active, a footer is appended:

---
[output truncated to {maxLines} lines – use `tool_output.per_tool.<tool>` to adjust]

Summarization Settings

Control how tool outputs are summarized for LLM context.

{
  "summaries": {
    "threshold_bytes": 102400,
    "exempt_tools": ["retrieve_summary", "task", "read"]
  }
}

• threshold_bytes – Output size threshold in bytes before summarization is triggered (default 102400 = 100KB).
• exempt_tools – Tools whose outputs are never summarized. Defaults to ["retrieve_summary", "task", "read"] to prevent re-summarization loops.

Note: The retrieve_summary tool supports paginated retrieval via offset and limit parameters to fetch large summarized outputs in chunks.

Disabling Agents

{
  "sme": { "disabled": true },
  "designer": { "disabled": true },
  "test_engineer": { "disabled": true }
}

Supported Languages

Full Tier-1 support: TypeScript, JavaScript, Python, Go, Rust
Tier-2 support: Java, Kotlin, C#, C/C++, Swift
Tier-3 support: Dart, Ruby, PHP/Laravel

All binaries optional. Missing tools produce soft warnings, never hard-fail.

Testing

6,000+ tests. Unit, integration, adversarial, and smoke. Run with:

bun test

Design Principles

1. Plan before code. Critic approves the plan before a single line is written.
2. One task at a time. Coder gets one task and full context. Nothing else.
3. Review everything immediately. Correctness, security, tests, adversarial tests. Every task.
4. Different models catch different bugs. Blind spots of the coder are the reviewer's strength.
5. Save everything to disk. Resume any project any day from .swarm/ state.
6. Document failures. Rejections and retries recorded. After 5 failures, escalate to you.

Documentation

• Getting Started — 15-minute first-run guide
• Documentation Index — navigate all docs
• Installation Guide — comprehensive reference
• Architecture Deep Dive — control model, pipeline, tools
• Design Rationale — why every major decision
• Commands Reference — all 41 /swarm subcommands
• Modes Guide — session modes (Turbo, Full-Auto) and project modes (strict/balanced/fast)
• Configuration — all config keys and examples
• Planning Guide — task format, phase structure, sizing

License

MIT

Stop hoping your agents figure it out. Start shipping code that actually works.