agent-task-splitter

Name: agent-task-splitter
Author: WenyuChiou

16stars4forks

How to install agent-task-splitter?

wenyuchiou/agent-collab-skills/agent-task-splitter

$npx -y skills add WenyuChiou/agent-collab-skills --skill agent-task-splitter

Installs into the current project.

Summary

Use when the user asks to split a goal across Claude, Codex, or Gemini (gemini requests are rerouted - lane deprecated); plan a multi-agent run; break work into parallel agent tasks; or decompose a large task that needs bounded context handoffs. This is the generic multi-agen

Files · 1

View on GitHub

SKILL.md

1	# agent-task-splitter
2
3	Bridge between a high-level goal and **the multi-agent execution
4	pipeline**. You write `.coord/plan.yml` (the DAG) and the per-agent
5	task files. The delegate skills (`codex-delegate`; historically
6	`gemini-delegate`) invoke the agents using those task files. The
7	reconciler reads what they produce.
8
9	This skill does not invoke any agent. It only plans and writes
10	files.
11
12	> **DEPRECATION + REROUTE (2026-06-18, updated 2026-07-10): the Gemini
13	> lane is DEAD — it fails closed. Never emit `agent: gemini` tasks.**
14	> Reroute what used to go there:
15	>
16	> \| Used to route to `gemini` \| Route now \|
17	> \|---\|---\|
18	> \| CJK / bilingual judgment, 語感, long-form writing \| `claude` (inline — judgment stays with the orchestrator) \|
19	> \| Bulk mechanical CJK (mirror sync, term sweeps) \| `codex` \|
20	> \| Long-context reading + synthesis \| `claude` inline, or `claude-cheap` when the reading is extraction/transcription-shaped \|
21	> \| Second-opinion review of generated output \| `claude` (a review is an honesty-critical task — never a cheap tier) \|
22	> \| Antigravity CLI (`agy`) via `antigravity-delegate` \| PROMOTED 2026-07-11: pre-registered k=5 reliability gate 5/5 (mc12: fresh sandbox per trial, decoy untouched, judgment question escalated verbatim, zero git ops) on top of the n=1 capability probe (mc11, 2026-07-10) — routable for bounded mechanical subtasks like `codex` / `claude-cheap`; cheap-tier guardrails unchanged (never reviews, completion verdicts, governance, or anything ambiguous) \|
23	>
24	> The `gemini` value remains PARSE-ONLY so reconcilers can read
25	> historical plans; §6b is retained as a legacy reference.
26
27	## Why use this instead of hand-rolling briefs
28
29	The supervisor (Claude) writing 2-3 brief files by hand looks cheap but
30	costs token + drift in two specific ways:
31
32	\| Hand-rolled briefs \| This skill \|
33	\|---\|---\|
34	\| Same context block (file paths, conventions, in-scope list) repeated across 3 task files — 3× redundant token cost in main session. \| Splitter writes once into `.coord/plan.yml`, references it from each task file. \|
35	\| Subtle drift between briefs ("Codex was told to update README, Gemini was told to mirror Stage 6" — but Codex's scope quietly included Stage 6 too). F11 incident shipped because of this. \| Splitter computes the file-scope set once and propagates the disjoint partition. \|
36	\| Operator forgets which agent gets which task type (codex for mechanical, gemini for long-context CJK). \| Routing rules baked in — same agent gets the same shape of work every time. \|
37
38	### Measured impact (real dogfood, 2026-05-14)
39
40	\| Setup \| Main session tokens \| Notes \|
41	\|---\|---\|---\|
42	\| With splitter (R2 + R4 combined: 2 parallel Codex + 1 mirror sync Gemini) \| ~9k tokens (R2 ~5k + R4 ~4k, measured) \| Splitter wrote 5 KB of plan.yml + briefs; main session read only structured summaries \|
43	\| Hand-rolled equivalent (estimated counterfactual) \| ~107-127k tokens (R2 control ~37k + R4 control ~70-90k) \| Operator inlines all context per brief, parses each agent's raw stdout, reconciles by hand \|
44	\| Saving \| ~12-14× combined (R2 ~7× + R4 ~17-22× per round) \| Plus the splitter's disjoint-scope partition prevented F11-class drift \|
45
46	The skill earns its keep when ≥ 2 subtasks go to different agents (or
47	same agent in parallel). For 1-shot delegation, call the delegate skill
48	directly.
49
50	### Anti-patterns this skill prevents
51
52	- F11 (cross-agent scope creep): Agent A sweeps a rule into files
53	that were Agent B's responsibility. Prevented by `.coord/plan.yml`'s
54	explicit `files_in_scope` partition per task.
55	- F14 (skipping the splitter for "small enough" 2-agent runs):
56	Operator decides to hand-roll because "it's only 2 tasks", and the
57	drift catches them later. The CLAUDE.md template below makes the
58	trigger mechanical: ≥ 2 parallel delegates → splitter is mandatory,
59	no judgment call.
60
61	### CLAUDE.md snippet to enforce routing
62
63	```markdown
64	## Multi-agent routing rule (enforced)
65
66	If a single round needs ≥ 2 delegate agents running in parallel (e.g.,
67	codex + claude-cheap, or 2 codex on independent subtasks): invoke
68	`Skill("agent-collab-workspace:agent-task-splitter", args="round=N ...")`
69	FIRST. Do NOT hand-roll briefs into `.ai/codex_task_*.md` directly when
70	≥ 2 are needed in the same round.
71
72	Decision rule (no judgment): 1 delegate per round → call delegate
73	dire

People who liked this also viewed

6 picks

Type

agent-task-splitter

16stars4forks

View on GitHub

How to install agent-task-splitter?

wenyuchiou/agent-collab-skills/agent-task-splitter

$npx -y skills add WenyuChiou/agent-collab-skills --skill agent-task-splitter

Installs into the current project.

Summary

Files · 1

View on GitHub

SKILL.md

1	# agent-task-splitter
2
3	Bridge between a high-level goal and **the multi-agent execution
4	pipeline**. You write `.coord/plan.yml` (the DAG) and the per-agent
5	task files. The delegate skills (`codex-delegate`; historically
6	`gemini-delegate`) invoke the agents using those task files. The
7	reconciler reads what they produce.
8
9	This skill does not invoke any agent. It only plans and writes
10	files.
11
12	> **DEPRECATION + REROUTE (2026-06-18, updated 2026-07-10): the Gemini
13	> lane is DEAD — it fails closed. Never emit `agent: gemini` tasks.**
14	> Reroute what used to go there:
15	>
16	> \| Used to route to `gemini` \| Route now \|
17	> \|---\|---\|
18	> \| CJK / bilingual judgment, 語感, long-form writing \| `claude` (inline — judgment stays with the orchestrator) \|
19	> \| Bulk mechanical CJK (mirror sync, term sweeps) \| `codex` \|
20	> \| Long-context reading + synthesis \| `claude` inline, or `claude-cheap` when the reading is extraction/transcription-shaped \|
21	> \| Second-opinion review of generated output \| `claude` (a review is an honesty-critical task — never a cheap tier) \|
22	> \| Antigravity CLI (`agy`) via `antigravity-delegate` \| PROMOTED 2026-07-11: pre-registered k=5 reliability gate 5/5 (mc12: fresh sandbox per trial, decoy untouched, judgment question escalated verbatim, zero git ops) on top of the n=1 capability probe (mc11, 2026-07-10) — routable for bounded mechanical subtasks like `codex` / `claude-cheap`; cheap-tier guardrails unchanged (never reviews, completion verdicts, governance, or anything ambiguous) \|
23	>
24	> The `gemini` value remains PARSE-ONLY so reconcilers can read
25	> historical plans; §6b is retained as a legacy reference.
26
27	## Why use this instead of hand-rolling briefs
28
29	The supervisor (Claude) writing 2-3 brief files by hand looks cheap but
30	costs token + drift in two specific ways:
31
32	\| Hand-rolled briefs \| This skill \|
33	\|---\|---\|
34	\| Same context block (file paths, conventions, in-scope list) repeated across 3 task files — 3× redundant token cost in main session. \| Splitter writes once into `.coord/plan.yml`, references it from each task file. \|
35	\| Subtle drift between briefs ("Codex was told to update README, Gemini was told to mirror Stage 6" — but Codex's scope quietly included Stage 6 too). F11 incident shipped because of this. \| Splitter computes the file-scope set once and propagates the disjoint partition. \|
36	\| Operator forgets which agent gets which task type (codex for mechanical, gemini for long-context CJK). \| Routing rules baked in — same agent gets the same shape of work every time. \|
37
38	### Measured impact (real dogfood, 2026-05-14)
39
40	\| Setup \| Main session tokens \| Notes \|
41	\|---\|---\|---\|
42	\| With splitter (R2 + R4 combined: 2 parallel Codex + 1 mirror sync Gemini) \| ~9k tokens (R2 ~5k + R4 ~4k, measured) \| Splitter wrote 5 KB of plan.yml + briefs; main session read only structured summaries \|
43	\| Hand-rolled equivalent (estimated counterfactual) \| ~107-127k tokens (R2 control ~37k + R4 control ~70-90k) \| Operator inlines all context per brief, parses each agent's raw stdout, reconciles by hand \|
44	\| Saving \| ~12-14× combined (R2 ~7× + R4 ~17-22× per round) \| Plus the splitter's disjoint-scope partition prevented F11-class drift \|
45
46	The skill earns its keep when ≥ 2 subtasks go to different agents (or
47	same agent in parallel). For 1-shot delegation, call the delegate skill
48	directly.
49
50	### Anti-patterns this skill prevents
51
52	- F11 (cross-agent scope creep): Agent A sweeps a rule into files
53	that were Agent B's responsibility. Prevented by `.coord/plan.yml`'s
54	explicit `files_in_scope` partition per task.
55	- F14 (skipping the splitter for "small enough" 2-agent runs):
56	Operator decides to hand-roll because "it's only 2 tasks", and the
57	drift catches them later. The CLAUDE.md template below makes the
58	trigger mechanical: ≥ 2 parallel delegates → splitter is mandatory,
59	no judgment call.
60
61	### CLAUDE.md snippet to enforce routing
62
63	```markdown
64	## Multi-agent routing rule (enforced)
65
66	If a single round needs ≥ 2 delegate agents running in parallel (e.g.,
67	codex + claude-cheap, or 2 codex on independent subtasks): invoke
68	`Skill("agent-collab-workspace:agent-task-splitter", args="round=N ...")`
69	FIRST. Do NOT hand-roll briefs into `.ai/codex_task_*.md` directly when
70	≥ 2 are needed in the same round.
71
72	Decision rule (no judgment): 1 delegate per round → call delegate
73	dire

People who liked this also viewed

6 picks

Type