agent-afk 3.37.0 → 3.38.0

package/dist/index.mjs

@@ -935,7 +935,7 @@ Subagents return their final assistant message verbatim \u2014 instruct them exp
 
 Foreground vs. background: by default (mode="foreground") this tool waits for the subagent to finish and returns its final message. Pass mode="background" to fire-and-forget \u2014 the tool returns a jobId immediately so you can keep working in the same turn. Background results are NOT auto-injected; retrieve them with the \`/bgsub:join <jobId>\` slash command (user surface) or by asking the user to join. Use background mode for long investigations the user does not need to wait on; use foreground for anything whose result you need to reason about in the same turn.
 
-Do not use this tool for: trivial one-file edits, conversational answers, direct tool calls the user explicitly requested, or tasks where dispatch overhead exceeds the work.`,input_schema:{type:"object",properties:{prompt:{type:"string",description:"The task for the agent to perform."},model:{type:"string",description:"Model for the agent. Defaults to parent session model. Override per-call to right-size cost vs. capability \u2014 `haiku` (cheapest/fastest), `sonnet` (general-use), `opus` (most capable). Append `_1m` (e.g. `sonnet_1m`) for 1M-context variants. Full model IDs are also accepted."},max_turns:{type:"number",description:"Maximum conversation turns (default 10, max 50)."},id_prefix:{type:"string",description:"Label prefix for log correlation."},mode:{type:"string",enum:["foreground","background"],description:'Execution mode. "foreground" (default) waits for the subagent to finish and returns its output. "background" returns a jobId immediately and leaves the subagent running detached \u2014 its result must be joined explicitly via /bgsub:join and is never auto-injected into this context. Background jobs are cancelled when the parent session ends.'}},required:["prompt"]}},ve={name:"skill",category:"skill",concurrencySafe:!1,description:"Invoke a registered skill by name. Skills are specialized capabilities that dispatch subagents with domain-specific prompts. Check the system prompt for the list of available skills and their descriptions.",input_schema:{type:"object",properties:{name:{type:"string",description:'Skill name (e.g., "mint", "diagnose", "shadow-verify").'},arguments:{type:"string",description:"Arguments to pass to the skill."}},required:["name"]}},Ee={name:"compose",category:"dag",concurrencySafe:!0,description:`Execute multiple subagent tasks as a DAG (directed acyclic graph). Nodes with no dependencies run in parallel; nodes with edges wait for their upstream dependencies to complete. Use when you need to orchestrate independent or dependent subagent work in a single call \u2014 e.g., diagnose in parallel with a fix, or research \u2192 implement \u2192 verify as a pipeline.
+Do not use this tool for: trivial one-file edits, conversational answers, direct tool calls the user explicitly requested, or tasks where dispatch overhead exceeds the work.`,input_schema:{type:"object",properties:{prompt:{type:"string",description:"The task for the agent to perform."},model:{type:"string",description:"Model for the agent. Defaults to parent session model. Override per-call to right-size cost vs. capability \u2014 `haiku` (cheapest/fastest), `sonnet` (general-use), `opus` (most capable). Append `_1m` (e.g. `sonnet_1m`) for 1M-context variants. Full model IDs are also accepted."},max_turns:{type:"number",description:"Maximum conversation turns (default 10, max 50)."},id_prefix:{type:"string",description:"Label prefix for log correlation."},mode:{type:"string",enum:["foreground","background"],description:'Execution mode. "foreground" (default) waits for the subagent to finish and returns its output. "background" returns a jobId immediately and leaves the subagent running detached \u2014 its result must be joined explicitly via /bgsub:join and is never auto-injected into this context. Background jobs are cancelled when the parent session ends.'},cwd:{type:"string",description:"Optional absolute path for the subagent to run in. When omitted, the child inherits the parent's working directory (e.g. an `afk -w` worktree). When provided, the child's file/shell tools (bash, grep, glob, read_file, write_file, edit_file) anchor at this path instead. Use to dispatch a subagent into a pre-existing git worktree you created with `bash: git worktree add <path>` so the subagent can work in isolation from the parent. Must be absolute (no relative paths) and must not contain `..` segments. Existence is not checked at dispatch time \u2014 a non-existent path surfaces as an error on the child's first cwd-relative tool call. Does not auto-propagate to further nested subagents \u2014 each `agent` call must specify `cwd` explicitly to operate in a worktree."}},required:["prompt"]}},ve={name:"skill",category:"skill",concurrencySafe:!1,description:"Invoke a registered skill by name. Skills are specialized capabilities that dispatch subagents with domain-specific prompts. Check the system prompt for the list of available skills and their descriptions.",input_schema:{type:"object",properties:{name:{type:"string",description:'Skill name (e.g., "mint", "diagnose", "shadow-verify").'},arguments:{type:"string",description:"Arguments to pass to the skill."}},required:["name"]}},Ee={name:"compose",category:"dag",concurrencySafe:!0,description:`Execute multiple subagent tasks as a DAG (directed acyclic graph). Nodes with no dependencies run in parallel; nodes with edges wait for their upstream dependencies to complete. Use when you need to orchestrate independent or dependent subagent work in a single call \u2014 e.g., diagnose in parallel with a fix, or research \u2192 implement \u2192 verify as a pipeline.
 
 Each node is a subagent task with its own prompt and optional model. Edges declare "from must finish before to starts." Omit edges entirely for pure parallel fan-out.