Subagent delegation
Core principle
Match each bounded assignment to the capability and judgment it needs, while
keeping architecture, integration, authorization, and final acceptance with
the lead. Installed agent definitions are authoritative for pinned settings.
Procedure
- Apply this playbook when the user invokes it or the user's instructions
direct its use for delegation; either covers delegation and reassignment
throughout the current task. Without either, a generic request to implement
work or use subagents does not apply it. Keep required constraints in agent
briefs so they apply without child agents loading this skill. Loading the
skill does not authorize implementation.
- Prefer solo work for small or tightly coupled tasks; use delegation or
worktrees when independent work or isolation justifies them. Leave task-graph
execution, worktree lifecycle, integration sequencing, and implementation
gates to the calling workflow, such as
implement-with-subagents.
- Choose the capability before the runtime role name: specific discovery,
predetermined mechanical checking, implementation with self-review and
repair ownership, or independent read-only analysis. For analysis, state
the mode: review against requirements, or investigation of a concrete
unresolved question. Establish concrete evidence before an investigation.
- Load only the active runtime's reference:
- Codex: exposed roles, installed TOML pins, model
and effort routing, and fresh-context calls.
- Claude Code: agent definitions, model and
effort routing, invocation overrides, and owner/reviewer lifecycle.
- Other runtimes: inspect their exposed capabilities and installed policy;
preserve configured choices rather than translating either provider's
model identifiers or effort values. Resolve missing required settings
before dispatch.
- Reassess capability and judgment for every new assignment, including
reassignment, independently of the parent task's size or model. Keep the
lead's model unchanged. File count, component count, or a sensitive area
alone does not raise a tier; when judgment matters, prefer a stronger model
at lower effort over a weaker model at higher effort. Preserve pinned
settings and access restrictions. If required capability or settings
are unavailable, report and resolve the mismatch before dispatch. Name the
assignment category in the brief and justify departures from the default
row. For investigations, name the unresolved question; for reviews,
identify the correctness checks; for synthesis, identify the evidence or
deliverable demands. At opus (Claude Code), Sol (Codex), or an equivalent
tier with high effort or above, make that question or those checks
concrete; touching a risky area alone does not qualify. If work stalls,
diagnose missing inputs, tool failures, or routine test repairs before
considering a higher tier.
- Give each agent a self-contained brief that fills every field of the
handoff brief. Tell
workers they share the codebase, must preserve others' edits, and must
accommodate concurrent changes. Further delegation
requires the lead's explicit authority. Keep remote mutations with the
root task holding the user's authorization, with one mutation owner.
- Launch independent assignments in parallel. Queue work behind another agent
only for a stated dependency, such as shared files, a contract, or ordering;
name that dependency. Separate worktrees alone do not remove dependencies.
- Choose independent audits proportionally to risk unless the calling
workflow requires one. Use a fresh read-only
reviewer without the implementation conversation. Supply original
requirements, the plan when available, and exact review scope; require a
ship, fix-first, or rethink verdict with evidence. Keep repairs with
the original worker and context where possible; the lead may take ownership
or explicitly reassign them.
- Inspect delegated diffs and verification evidence before accepting the work.
Repeat checks when evidence is incomplete or affected code changes, without
duplicating adequate validation. Return failures to the owner for repair
and reassess affected evidence.
Finish gate
Finish when the lead has inspected the assigned result and its evidence,
resolved failures or reported the specific blocker, and made the acceptance
decision under the calling workflow. For advice-only requests, finish with the
assignment choices and their reasons; do not spawn agents or start execution.
1---2name: subagent-delegation3description: Use when the user invokes the subagent delegation playbook, or the user's instructions direct its use when delegating work or deciding whether to delegate.4metadata:5 source: https://skillmd.com/skills/tuyv/subagent-delegation6---78# Subagent delegation910## Core principle1112Match each bounded assignment to the capability and judgment it needs, while13keeping architecture, integration, authorization, and final acceptance with14the lead. Installed agent definitions are authoritative for pinned settings.1516## Procedure17181. Apply this playbook when the user invokes it or the user's instructions19 direct its use for delegation; either covers delegation and reassignment20 throughout the current task. Without either, a generic request to implement21 work or use subagents does not apply it. Keep required constraints in agent22 briefs so they apply without child agents loading this skill. Loading the23 skill does not authorize implementation.242. Prefer solo work for small or tightly coupled tasks; use delegation or25 worktrees when independent work or isolation justifies them. Leave task-graph26 execution, worktree lifecycle, integration sequencing, and implementation27 gates to the calling workflow, such as `implement-with-subagents`.283. Choose the capability before the runtime role name: specific discovery,29 predetermined mechanical checking, implementation with self-review and30 repair ownership, or independent read-only analysis. For analysis, state31 the mode: review against requirements, or investigation of a concrete32 unresolved question. Establish concrete evidence before an investigation.334. Load only the active runtime's reference:34 - [Codex](references/codex.md): exposed roles, installed TOML pins, model35 and effort routing, and fresh-context calls.36 - [Claude Code](references/claude-code.md): agent definitions, model and37 effort routing, invocation overrides, and owner/reviewer lifecycle.38 - Other runtimes: inspect their exposed capabilities and installed policy;39 preserve configured choices rather than translating either provider's40 model identifiers or effort values. Resolve missing required settings41 before dispatch.425. Reassess capability and judgment for every new assignment, including43 reassignment, independently of the parent task's size or model. Keep the44 lead's model unchanged. File count, component count, or a sensitive area45 alone does not raise a tier; when judgment matters, prefer a stronger model46 at lower effort over a weaker model at higher effort. Preserve pinned47 settings and access restrictions. If required capability or settings48 are unavailable, report and resolve the mismatch before dispatch. Name the49 assignment category in the brief and justify departures from the default50 row. For investigations, name the unresolved question; for reviews,51 identify the correctness checks; for synthesis, identify the evidence or52 deliverable demands. At opus (Claude Code), Sol (Codex), or an equivalent53 tier with high effort or above, make that question or those checks54 concrete; touching a risky area alone does not qualify. If work stalls,55 diagnose missing inputs, tool failures, or routine test repairs before56 considering a higher tier.576. Give each agent a self-contained brief that fills every field of the58 [handoff brief](references/subagent-selection.md#handoff-brief). Tell59 workers they share the codebase, must preserve others' edits, and must60 accommodate concurrent changes. Further delegation61 requires the lead's explicit authority. Keep remote mutations with the62 root task holding the user's authorization, with one mutation owner.637. Launch independent assignments in parallel. Queue work behind another agent64 only for a stated dependency, such as shared files, a contract, or ordering;65 name that dependency. Separate worktrees alone do not remove dependencies.668. Choose independent audits proportionally to risk unless the calling67 workflow requires one. Use a fresh read-only68 reviewer without the implementation conversation. Supply original69 requirements, the plan when available, and exact review scope; require a70 `ship`, `fix-first`, or `rethink` verdict with evidence. Keep repairs with71 the original worker and context where possible; the lead may take ownership72 or explicitly reassign them.739. Inspect delegated diffs and verification evidence before accepting the work.74 Repeat checks when evidence is incomplete or affected code changes, without75 duplicating adequate validation. Return failures to the owner for repair76 and reassess affected evidence.7778## Finish gate7980Finish when the lead has inspected the assigned result and its evidence,81resolved failures or reported the specific blocker, and made the acceptance82decision under the calling workflow. For advice-only requests, finish with the83assignment choices and their reasons; do not spawn agents or start execution.