> ## Documentation Index > Fetch the complete documentation index at: https://docs.bastani.ai/llms.txt > Use this file to discover all available pages before exploring further. # Subagents > Run focused Atomic child agents # Subagents Atomic bundles `@bastani/subagents`, an extension for bounded specialist delegation with separate context while the parent remains in control. Use a single agent, chain, or parallel fan-out when isolation or a specialist pass materially helps with locating code, analyzing behavior, researching references, reproducing actual failures, or simplifying code. Keep interactive, exploratory, conceptual, and conversation-led work inline when direct user steering is more useful. You do not need to install anything separately when you use `@bastani/atomic`. ## Start with natural language Ask Atomic to coordinate subagents in plain language: ```text theme={null} Map the authentication flow with focused subagents before we change it. ``` ```text theme={null} Run a parallel review composition: one pass for current behavior, one for failure modes, and one for existing patterns. ``` ```text theme={null} Research the upstream library behavior online, then compare it with our local implementation. ``` Atomic decides whether delegation adds value, which specialist fits each bounded part, and whether the work should run as a single child, parallel group, chain, foreground run, or selective background run. Multiple steps, files, tests, validation, or parallelism alone do not require a workflow; clearly delegated long-running autonomous work that needs durable stages, checkpoints, resumability, HIL, gates, retries, or loops is usually better served by a workflow. ## Subagent execution is non-interactive Supported subagent launches start immediately without opening a preview/editor prompt or waiting for terminal input. This applies to single, parallel, chain, foreground, background, fanout, prompt-template, and human-entered `/run`, `/chain`, `/parallel`, and `/run-chain` execution. Ask any necessary questions in the parent conversation before delegating. The human slash commands remain registered and continue to use their separate parsing and event-bridge path, including background and fork flags. Subagents now run and return their results directly. Atomic does not infer acceptance gates from prompt wording, inject `acceptance-report` instructions into child prompts, parse or strip `acceptance-report` blocks, or reject completed child runs because changed-file, test, or review evidence is missing. Put any evidence or validation requirements directly in the task text you give the parent or child agent. ## Foreground supervisor coordination When a foreground child sends `intercom.ask`, `intercom.send`, or `contact_supervisor` coordination, Atomic first probes for the exact foreground owner. Only an exact live child reserves the request; Atomic then sends a generation-scoped detach commit and waits for that child to acknowledge it before placing the message in the parent's model-visible steering queue. This first-refusal ordering also applies when the parent is a busy workflow stage: detach completes before the request enters the stage AgentSession generation boundary, breaking the child-waits-for-reply / stage-waits-for-child cycle. Unmatched and background-child messages retain existing routing—ordinary parents queue until idle, while open workflow stages fall back to their native generation admission. Blocking `need_decision` and `interview_request` calls remain actionable through [Intercom](/intercom)'s pending/reply tracker, and the exact threaded reply resumes the retained child without delayed duplicate delivery. Only the matching foreground child can authorize release of the parent `subagent` tool. For a parallel foreground group, that accepted commit releases foreground supervision for every active sibling as one unit, so a long-running sibling cannot keep a blocking child request trapped behind the aggregate tool call; tasks still waiting behind the concurrency limit are skipped and never launched unsupervised. Active child processes stay alive under normal watchdog, cancellation, drain, and stdio cleanup ownership; each eventual completion replaces its detached placeholder. Run-owned worktrees remain attached until every detached child closes, then Atomic captures their diffs before cleanup. Fire-and-forget `intercom.send` and `progress_update` also release foreground supervision promptly, but do not create a reply waiter. Blocking coordination is race-safe: a session holds at most one outbound reply waiter, and concurrent blocking requests (parallel `intercom.ask` calls, or `intercom.ask` racing `contact_supervisor`) settle atomically. One request wins the reservation; every other concurrent call returns a normal "Already waiting for a reply" tool error without crashing the agent process or disturbing the pending ask. Cancellation and send failures release only their own waiter, and threaded replies still resolve the exact winning request. Subagent result announcements are also resilient in sessions that never receive an extension `session_start` (for example non-interactive in-process child sessions): the lazy Intercom runtime initializes from the most recent turn/tool lifecycle context and delivers self-addressed results locally. If no context is available at all, the relay acknowledges the announcement as undelivered — the `subagent` tool then falls back to returning results inline — instead of recording connection errors in the session transcript. Intercom connection remains tool-driven. Foreground and background launches do not import the heavy Intercom runtime or connect either the parent or bridged child automatically. If live child-to-parent coordination is needed, the parent model should invoke `intercom({ action: "status" })` before launch; the child then connects on its first `contact_supervisor` or `intercom` call. Cancellation or session replacement still invalidates the handshake generation, so stale acknowledgements cannot surface or detach a child. Atomic's implementation adapts the prompt foreground release and later-result recovery contracts proven in `nicobailon/pi-subagents` commits `1b55c8c`, `589e51e`, `68fb528`, and `9dfe3df`; it retains Atomic's broker and raw-TypeScript architecture rather than copying upstream's filesystem transport. ## Migration from acceptance gates If you have older subagent calls, saved chains, or custom agents that used the removed gate fields: * Remove `acceptance` properties from `subagent()` calls, `tasks` entries, `chain` steps, static parallel task items, and dynamic fanout parallel templates. Atomic no longer reads these fields; JSON chain rewrites drop legacy copies. * Remove `completionGuard: false` from agent frontmatter and custom agent definitions. The no-mutation completion guard no longer exists, so the override has no effect and management rewrites strip it. * Move validation, command, evidence, review, or residual-risk requirements into the natural-language task text passed to the parent or child agent. ## Bundled agents Atomic currently bundles these agents from `@bastani/subagents`: | Agent | Use it for | Edit files? | | ---------------------------- | ------------------------------------------------------------------------------------- | ------------------------ | | `codebase-locator` | Find relevant files, directories, tests, configs, and docs for a topic. | No | | `codebase-analyzer` | Explain how specific code works and trace data flow with file references. | No | | `codebase-pattern-finder` | Find similar implementations, conventions, and test examples to model after. | No | | `codebase-research-locator` | Locate prior `research/` and `specs/` documents related to the task. | No | | `codebase-research-analyzer` | Extract decisions, constraints, and still-relevant conclusions from prior local docs. | No | | `codebase-online-researcher` | Research official docs, ecosystem behavior, and open-source source references online. | Can write research notes | | `debugger` | Reproduce, diagnose, and fix failures or unexpected behavior. | Yes | | `code-simplifier` | Clean up recently changed code while preserving behavior. | Yes | Read-oriented agents should inspect and report. `debugger` and `code-simplifier` can edit files, so run them with an explicit scope and validation target. ## Review compositions Atomic does not bundle a single generic review agent. Instead, compose specialists with distinct angles and let the parent session synthesize their findings before applying any fix. Common review angles: | Angle | Specialist pattern | | ----------------------------------- | ------------------------------------------------------------------------------------------------------------------- | | Current behavior and regressions | `codebase-analyzer` inspects the changed flow and cites file/line evidence. | | Failure modes | `debugger` runs in inspect-only mode to reproduce or reason about likely failures without editing. | | Fit with project conventions | `codebase-pattern-finder` compares the patch with existing local examples. | | Prior decisions | `codebase-research-locator` finds relevant docs, then `codebase-research-analyzer` extracts applicable constraints. | | External API or library conformance | `codebase-online-researcher` checks authoritative sources and version-specific behavior. | Example request: ```text theme={null} Review the current diff with fresh-context specialists: analyze correctness, inspect failure modes without editing, and compare the implementation to existing patterns. Synthesize only issues worth fixing now. ``` Useful prompt templates include `/parallel-review`, `/review-loop`, `/parallel-research`, `/parallel-context-build`, `/parallel-handoff-plan`, and `/parallel-cleanup`. Treat them as reusable compositions, not as separate bundled agent names. ## Background work and control Foreground subagents stream progress in the conversation and are the right default when the parent needs the result before proceeding. Use background subagents selectively for genuinely long-running or independently useful bounded delegation; they keep working after control returns and report completion later. Natural-language examples: ```text theme={null} Run the local research scan in the background. ``` ```text theme={null} Show me the current async subagent runs. ``` Tool examples: ```ts theme={null} subagent({ agent: "codebase-analyzer", task: "Trace the auth flow with file references.", async: true }) subagent({ action: "status" }) subagent({ action: "status", id: "" }) subagent({ action: "interrupt", id: "" }) subagent({ action: "resume", id: "", message: "continue with the test failures" }) subagent({ action: "doctor" }) ``` Use `interrupt` when you want a resumable stop. Use `resume` to send a follow-up to a reachable async child, or to revive a completed child from its saved session when the run has enough metadata. Use `doctor` for read-only setup diagnostics. Background runs are detached. Their acknowledgement explicitly says the run was launched and completion is pending: the launch tool call itself is terminal, while the detached child continues and will notify the originating session when it completes. If Atomic has no useful independent work in the meantime, it should end the turn instead of polling in a loop. Completion delivery distinguishes two compatibility surfaces. Intercom delivery is confirmation-based and preserves a successful phase across watcher replacement, so another phase can retry without replaying the parent message. The in-process `subagent:async-complete` event remains a synchronous compatibility emission: returning without an explicit synchronous rejection counts as local acceptance even when no listener is installed. Equivalent result-file aliases coalesce by canonical run identity, while aliases that reuse that identity with different user-visible output or parent targets are retained under collision-resistant names in the non-scanned `.undelivered` directory instead of being delivered or deleted as duplicates. Modern results whose status is not terminal are rechecked with capped exponential delays and still recover if terminal status appears later. Delivery failures also back off; after a finite sequence of attempts with no phase progress, Atomic retains the still-owned result in `.undelivered` and logs its path rather than retrying forever or deleting the payload. Terminal background notifications preserve each child’s accepted Intercom order. Every broker message retains its sender timestamp and source session ID/name; bridged-child messages additionally carry their source run ID, agent, and child index. The deterministic child Intercom target and result-relay child mapping use the same run/agent/index identity, so duplicate display aliases remain isolated by run ownership. Before a paused, completed, or failed result is delivered, Intercom synchronously claims pre-terminal ordinary messages from those exact child targets and atomically admits the FIFO prelude plus terminal message as one ordered custom-message batch. A process-local bridge complements the extension event bus because lazily activated companion extensions may have distinct loader buses; duplicate bridge/event delivery is deduplicated by exact terminal identity even when the successful dispatch has no prelude. Failed dispatches remain retryable, and a resumed run’s later lifecycle terminal has a distinct identity. Terminal admission does not wait for a separate model turn, unrelated child entries remain independently queued, and exact threaded ask replies retain their correlation and timeout behavior. Inside workflow stages, completion delivery observes the stage generation boundary. A completion received before the boundary closes is queued through the stage AgentSession and processed before the stage publishes its terminal snapshot. A detached completion that arrives after close is routed once to the parent/main chat and cannot automatically reopen or append to the completed stage transcript. Producers that are still running do not hold the stage open, so background work remains non-blocking; explicit post-mortem stage chat is still available separately. When a workflow graph overlay is open, Atomic also publishes the live async subagent summary into the shared status surface. The below-editor async widget remains available when the workflow overlay is hidden, and the overlay statusline keeps the run count/state visible while the graph fills the terminal. ## Context and execution modes Subagents can run with fresh or forked context: * `context: "fresh"` starts a separate child with only the task and selected agent context. * `context: "fork"` creates a real branched child session from the parent session leaf. It fails fast if the parent session cannot be forked; it does not silently downgrade to fresh context. For adversarial review or research, prefer fresh context so the specialist inspects the repository directly. Use forked context when a writer needs the parent conversation history in a separate branch. For parallel implementation work, `worktree: true` can give each child an isolated git worktree so concurrent edits do not clobber each other. Fresh child processes use normal Atomic package discovery when an agent omits `extensions`, so bundled lightweight MCP, web-access, and Intercom wrappers are available just as they are in the parent. An explicit `extensions` field (including an empty list) intentionally switches the child to extension-allowlist mode and excludes unlisted builtins; it does not inherit the parent's normal discovery set. Top-level parallel calls support up to 50 subagents after expanding each task's optional `count`. The extension's `parallel.maxTasks` setting defaults to 50 and can enforce a lower task limit; `parallel.concurrency` independently controls how many of those children run at once. Subagent tasks, parallel/chain items, and the top-level call accept a `group` field that sets the spawned child's [Intercom](/intercom) home group, so same-group subagents can intercom each other while staying isolated from other groups. A named string joins that group; `true` auto-generates one shared UUID group per parallel set (shared across all items in that set). Precedence is `explicit subagent group > inherited current-session (stage) group > env ATOMIC_INTERCOM_GROUP > config > "default"`. When a subagent does not specify a group, it **inherits the launching session's group** — read race-safely from the session, not from global env — so subagents spawned by a grouped workflow stage automatically join that stage's group and can coordinate with the stage and its sibling subagents. The child intercom group is applied only when the child actually has intercom access (the peer `intercom` tool or the subagent-only `contact_supervisor` tool); a child without intercom is never placed into a group. `contact_supervisor` still reaches the supervisor across group boundaries because Atomic requests a broker capability before spawn and binds the child's registration to the issuing supervisor. Foreground and single-child paths use exact child scopes; asynchronous chains use bounded per-child slots so indexes shifted by dynamic fanout remain valid. Capability environment values are explicitly cleared for descendants without a fresh grant, and parent-held grants are restored after reconnects. The lightweight Intercom wrapper lazy-loads the authorization provider; provider failures propagate and abort launch, while hosts without a provider omit supervisor metadata instead of exposing a broken `contact_supervisor` channel. When a subagent call, parallel task, chain step, or background run uses a `cwd`, Atomic validates that working directory before starting the child runtime. Missing or non-directory paths are reported as `cwd` problems instead of lower-level process-spawn errors, so failures point at the requested child workspace rather than at the runtime binary. Single-agent calls also accept `reads: string[] | false`. Atomic prepends those files as read context for foreground and background execution through the same path resolver, including `/run agent[reads=a.md+b.md]`. Relative entries resolve against the effective child `cwd` (including a relative top-level `cwd` resolved from the parent); absolute entries are unchanged. Invalid values fail before either child runtime starts. Single-agent calls accept `progress: boolean` in foreground, background, and revived/resumed mode. `progress: true` creates a run-scoped `progress.md` under isolated subagent artifact storage and instructs the child to maintain it without writing `progress.md` into the child `cwd`; `progress: false` disables an agent's `defaultProgress`. When `progress` is omitted, the agent's default is inherited, except that inherited progress is suppressed for read-only tasks (`progress: true` still explicitly opts in). Foreground runs remove this run-owned progress storage after the child exits when `artifacts: false`, including children temporarily detached for intercom coordination. This is separate from `includeProgress: true`, which only includes detailed runtime progress telemetry in the final tool result and does not create or maintain a file. ```ts theme={null} subagent({ agent: "worker", task: "Implement the approved fix.", progress: true }) subagent({ agent: "worker", task: "Implement it in the background.", progress: true, async: true }) ``` ## Nested and fanout boundaries Child-safety boundaries are enforced by the bundled subagent extension: * Normal child sessions do not receive the `subagent` tool or the parent-only subagents skill. * Child context is filtered to remove parent orchestration artifacts, old control/status messages, and prior parent `subagent` tool calls/results. * Non-fanout children are instructed that they are not the parent orchestrator and must not propose or run subagents. * Nested fanout is available only for explicitly authorized agents whose resolved tools include `subagent`. Authorized fanout children receive narrower instructions that limit delegation to the assigned fanout. * The recursion guard defaults to a hard maximum of five delegated subagent levels. `ATOMIC_SUBAGENT_MAX_DEPTH`, extension `config.maxSubagentDepth`, and agent frontmatter can choose a lower value from `0` to `5`; higher values are clamped. This keeps the parent session responsible for orchestration unless you deliberately choose a fanout-capable custom agent. ## Custom agents Custom agents are Markdown files with YAML frontmatter and a system prompt body. Common locations are: | Scope | Path | | ------- | -------------------------------- | | User | `~/.atomic/agent/agents/**/*.md` | | Project | `.atomic/agents/**/*.md` | A small custom read-only inspection agent: ```markdown theme={null} --- name: strict-inspector description: Inspect code for correctness and regressions tools: read, search, bash model: anthropic/claude-sonnet-4 fallbackModels: openai/gpt-5-mini inheritProjectContext: true --- You are a read-only inspector. Inspect the current diff, cite evidence with file paths, and return only issues worth fixing now. Do not edit files. ``` If an agent or chain step uses an explicit empty `tools: []` allowlist together with `outputSchema`, Atomic starts the child with only `structured_output` enabled for the required final answer. It does not omit `--tools` and accidentally restore default tools. Path-only tool entries remain extension paths and do not create a builtin allowlist by themselves. The child prompt-runtime extension is loaded before user/tool extensions so its schema-backed `structured_output` tool is registered before explicit allowlists are applied. ## Structured output schemas Chain and parallel steps can declare an `outputSchema` when the parent needs reliable machine-readable handoff data. Atomic passes that schema directly to a `structured_output` tool backed by the shared Atomic factory. The child should call `structured_output` when it is done: ```ts theme={null} structured_output({ files: ["src/auth.ts"], risks: ["missing regression test"], }) ``` `outputSchema` is a plain JSON Schema descriptor object. It may describe object, array, or primitive final values, and the child should pass a JSON value that matches that schema directly. Atomic no longer adds object-root restrictions, sidecar metadata, transcript-finality checks, or duplicate-call guards. The child runtime writes the tool arguments to `output.json`; the parent validates that captured JSON against the schema, reads it back as `result.structuredOutput`, and exposes it in named-chain references under `outputs.name.structured`. If the child exits without calling `structured_output`, or the captured value fails schema validation, Atomic retries up to three times with a corrective prompt that quotes the exact contract/validation error and reminds the child to call `structured_output` rather than returning plain JSON. Children without `outputSchema` do not receive `structured_output` from Atomic's default tool registry. They can still use a custom extension-provided terminating tool if you explicitly add one. Dynamic fanout `collect.outputSchema` validates the collected result array after child runs finish. ## Fallback models Agents can define ordered `fallbackModels` for retryable provider or model failures such as rate limits, quota/usage-limit exhaustion (for example a provider reporting `The usage limit has been reached`, or `usage_limit_reached`/`insufficient_quota` codes), auth problems, unavailable models, network timeouts, or 5xx errors. Atomic tries the requested primary model first, then configured fallbacks, and finally appends the current user-selected model as the last fallback candidate when available. Cancellations, safety refusals, and task/tool failures are never retried on another model. A candidate that cannot serve the current request — for example an HTTP 400/413/422 bad/unprocessable/payload-too-large request, an unsupported tool or parameter, a context-length/context-window overflow, or a `too large` / `invalid_request` error — is treated as request/context incompatible and the chain advances to the next candidate rather than stopping. This means that if none of the configured candidates are applicable to the request, Atomic falls back to the currently selected user model instead of failing outright. Each foreground and background model candidate is bounded by a per-attempt idle watchdog (default 5 minutes without child stdout, stderr, or JSON child events) and an absolute wall-clock cap (default 60 minutes). An in-flight tool execution counts as activity, so a slow, quiet tool call (a long build or test run that streams nothing until it finishes) is not mistaken for a stalled attempt; only the wall-clock cap bounds such attempts. If either watchdog trips, Atomic terminates that child attempt, records a retryable timeout in `modelAttempts`, and continues to the next fallback candidate. The defaults can be overridden with `ATOMIC_SUBAGENT_ATTEMPT_IDLE_TIMEOUT_MS` and `ATOMIC_SUBAGENT_ATTEMPT_TIMEOUT_MS`; `ATOMIC_SUBAGENT_ATTEMPT_KILL_GRACE_MS` controls SIGTERM-to-SIGKILL escalation. Setting the idle or wall-clock variable to `0` (or a negative value) disables that timeout entirely; non-numeric values are ignored and the default applies. The kill-grace period cannot be disabled — `0`, negative, or non-numeric values fall back to its default so escalation always stays bounded. When registry availability shows that a known candidate provider has no configured auth, Atomic records a skipped model attempt before spawning a child. Unknown/custom providers are still attempted, and the current user-selected model appended as the final fallback is never filtered out by this pre-spawn check. Fallbacks do not retry ordinary task failures, validation failures, tool failures, cancellations, or workflow-code errors. Because a fallback may send the same prompt and context to a different provider, choose models that match your cost, privacy, and data-handling requirements. Each candidate can also carry its own reasoning effort — see [Reasoning levels](#reasoning-levels). ## Reasoning levels Set the reasoning (thinking) effort for each model candidate with a `model_name:thinking_effort` suffix on `model` and on every `fallbackModels` entry. Valid efforts are `off`, `minimal`, `low`, `medium`, `high`, `xhigh`, and `max` — the same shorthand used by `atomic --model sonnet:high`. `xhigh` and `max` are used only when the selected model's capability map supports them. ```markdown theme={null} --- name: deep-reviewer description: Adversarial reviewer for risky diffs tools: read, search, bash model: anthropic/claude-sonnet-4:high fallbackModels: openai/gpt-5:medium, anthropic/claude-haiku-4-5:off --- ``` Because the effort travels with each model string, every primary and fallback candidate is self-contained: a fallback can run at a different effort than the primary, so a high-effort primary degrades gracefully to a cheaper, lower-effort fallback. **Migrate off the legacy `thinking` field.** The separate `thinking:` frontmatter field is deprecated. It still works as a default for any candidate that has no suffix, and a suffix always wins, but new agents should encode the effort directly on `model` and `fallbackModels`: ```diff theme={null} -model: openai/gpt-5.5 -fallbackModels: anthropic/claude-opus-4-8 -thinking: xhigh +model: openai/gpt-5.5:xhigh +fallbackModels: anthropic/claude-opus-4-8:xhigh ``` `fallbackThinkingLevels` exists only as an optional compatibility helper: it is aligned by index to `fallbackModels` and supplies a fallback candidate's effort only when that fallback entry has no suffix. Prefer suffixed model strings instead. Attempt metadata reports the resolved model and the effective reasoning effort used for each attempt. ## Related docs * [Workflows](/workflows) for multi-stage reusable automation. * [Intercom](/intercom) for cross-session messaging and supervisor escalation. * [Skills](/skills) for reusable instructions invoked with `/skill:`. * [Settings](/settings) for user and project configuration.