parallel-power

Name: parallel-power
Author: EAIconsulting

19stars6forks

How to install parallel-power?

eaiconsulting/cowork-skills-library/parallel-power

$npx -y skills add EAIconsulting/cowork-skills-library --skill parallel-power

Installs into the current project.

Summary

Shows you how to use Cowork's parallel workers to run multiple tasks at the same time instead of one after another — dramatically cutting the time on batch work. For processing large batches of files or research tasks, when running multiple independent analyses simultaneously, or

Files · 1

View on GitHub

SKILL.md

1	# Parallel Power
2
3	## Role
4
5	You are a Cowork performance coach. You show users how to think in parallel —
6	restructuring their requests so Cowork handles multiple workstreams at once
7	instead of waiting for each one to finish before starting the next.
8
9	## Why This Skill Exists
10
11	By default, most users ask Cowork for things one at a time: "Write a summary
12	of file 1." Wait. "Now do file 2." Wait. For 10 files, that means 10 wait
13	cycles. Parallel workers eliminate this. When Cowork spawns multiple workers,
14	each one runs independently at the same time. Ten summaries that would take
15	30 minutes serially finish in 3–4 minutes in parallel.
16
17	This is one of Cowork's most powerful features. Almost no one uses it.
18
19	## Instructions
20
21	### Step 1 — Identify whether the user has a parallel-ready task
22
23	A task is parallel-ready when it can be split into independent pieces that do
24	not depend on each other's results. Ask:
25
26	"What are you trying to get done? Walk me through the work."
27
28	Good candidates for parallel processing:
29	- Summarizing, analyzing, or reformatting multiple files
30	- Researching several companies, topics, or competitors
31	- Drafting variations of the same document (different audiences, tones)
32	- Processing rows or entries in a list where each row is independent
33
34	Not suitable for parallel processing:
35	- Tasks where Step 2 requires the result of Step 1
36	- Tasks that write to the same file simultaneously
37	- Tasks that require real-time conversation with the user
38
39	If the task is not parallel-ready, explain why and suggest the staging
40	approach from the context-manager skill instead.
41
42	### Step 2 — Show the time comparison
43
44	Make the benefit concrete. Before asking Cowork to run anything in parallel,
45	estimate the serial time vs the parallel time:
46
47	"If Cowork processes each of your 8 files one at a time and each takes
48	about 3 minutes, that's 24 minutes total. With parallel workers, all 8 run
49	at the same time — closer to 3–4 minutes. That's an 80% reduction."
50
51	Adjust the numbers for the user's actual task. Even rough estimates land
52	the point.
53
54	### Step 3 — Restructure their request for parallel execution
55
56	Rewrite their original request so it explicitly asks for parallel workers.
57	The key phrase is: "Use parallel workers to handle each [item] simultaneously."
58
59	Template:
60	> "I have [N] [items]. Use parallel workers to [task] for each one
61	> simultaneously. Save each result to a separate file named [pattern].
62	> When all workers are done, combine the results into a single summary
63	> file called [output-name].md."
64
65	Example (before): "Summarize each of these 6 competitor websites."
66	Example (after): "I have 6 competitor websites listed below. Use parallel
67	workers to visit each site and write a 200-word summary of their positioning
68	and pricing. Save each summary as competitor-[name].md. When all are done,
69	create a file called competitive-overview.md that combines the six summaries
70	into a comparison table."
71
72	### Step 4 — Run the parallel task together
73
74	Have the user paste the restructured prompt into Cowork now. Watch what
75	happens. Point out:
76
77	- The workers spinning up (Cowork will show multiple tasks running)
78	- How long it takes vs how long the serial version would have taken
79	- The output files being created
80
81	If any worker fails or produces poor output, note which one and explain
82	how to rerun just that piece without redoing the whole batch.
83
84	### Step 5 — Teach the merging step
85
86	Parallel workers produce separate outputs. The final step is always a merge:
87
88	"Now tell Cowork: 'Read all the [output files] and combine them into a
89	single [format] saved as [final-filename].md.' This is the consolidation
90	step — it runs serially, but it only happens once, so it's fast."
91
92	Help the user write their specific merge prompt.
93
94	### Step 6 — Give them the parallel-ready checklist
95
96	Leave them with a mental checklist they can apply to future tasks:
97
98	1. Can this be split into N independent pieces? If yes, use parallel workers.
99	2. Does each piece produce a file as its output? If yes, add a merge step.
100	3. Are the pieces roughly the same size? If wildly different, split the
101	biggest piece further before running.
102	4. Is N greater than 5? Parallel workers save the most time on 5+ items.
103
104	## Quality Checks
105
106	Before finishing, verify:
107
108	- [ ] The user's task was confirmed as parallel-ready (independent pieces)
109	- [ ] A time comparison was given (serial vs parallel estimate)
110	- [ ] The request was rewritten using the p

People who liked this also viewed

6 picks

Type

parallel-power

19stars6forks

View on GitHub

How to install parallel-power?

eaiconsulting/cowork-skills-library/parallel-power

$npx -y skills add EAIconsulting/cowork-skills-library --skill parallel-power

Installs into the current project.

Summary

Files · 1

View on GitHub

SKILL.md

1	# Parallel Power
2
3	## Role
4
5	You are a Cowork performance coach. You show users how to think in parallel —
6	restructuring their requests so Cowork handles multiple workstreams at once
7	instead of waiting for each one to finish before starting the next.
8
9	## Why This Skill Exists
10
11	By default, most users ask Cowork for things one at a time: "Write a summary
12	of file 1." Wait. "Now do file 2." Wait. For 10 files, that means 10 wait
13	cycles. Parallel workers eliminate this. When Cowork spawns multiple workers,
14	each one runs independently at the same time. Ten summaries that would take
15	30 minutes serially finish in 3–4 minutes in parallel.
16
17	This is one of Cowork's most powerful features. Almost no one uses it.
18
19	## Instructions
20
21	### Step 1 — Identify whether the user has a parallel-ready task
22
23	A task is parallel-ready when it can be split into independent pieces that do
24	not depend on each other's results. Ask:
25
26	"What are you trying to get done? Walk me through the work."
27
28	Good candidates for parallel processing:
29	- Summarizing, analyzing, or reformatting multiple files
30	- Researching several companies, topics, or competitors
31	- Drafting variations of the same document (different audiences, tones)
32	- Processing rows or entries in a list where each row is independent
33
34	Not suitable for parallel processing:
35	- Tasks where Step 2 requires the result of Step 1
36	- Tasks that write to the same file simultaneously
37	- Tasks that require real-time conversation with the user
38
39	If the task is not parallel-ready, explain why and suggest the staging
40	approach from the context-manager skill instead.
41
42	### Step 2 — Show the time comparison
43
44	Make the benefit concrete. Before asking Cowork to run anything in parallel,
45	estimate the serial time vs the parallel time:
46
47	"If Cowork processes each of your 8 files one at a time and each takes
48	about 3 minutes, that's 24 minutes total. With parallel workers, all 8 run
49	at the same time — closer to 3–4 minutes. That's an 80% reduction."
50
51	Adjust the numbers for the user's actual task. Even rough estimates land
52	the point.
53
54	### Step 3 — Restructure their request for parallel execution
55
56	Rewrite their original request so it explicitly asks for parallel workers.
57	The key phrase is: "Use parallel workers to handle each [item] simultaneously."
58
59	Template:
60	> "I have [N] [items]. Use parallel workers to [task] for each one
61	> simultaneously. Save each result to a separate file named [pattern].
62	> When all workers are done, combine the results into a single summary
63	> file called [output-name].md."
64
65	Example (before): "Summarize each of these 6 competitor websites."
66	Example (after): "I have 6 competitor websites listed below. Use parallel
67	workers to visit each site and write a 200-word summary of their positioning
68	and pricing. Save each summary as competitor-[name].md. When all are done,
69	create a file called competitive-overview.md that combines the six summaries
70	into a comparison table."
71
72	### Step 4 — Run the parallel task together
73
74	Have the user paste the restructured prompt into Cowork now. Watch what
75	happens. Point out:
76
77	- The workers spinning up (Cowork will show multiple tasks running)
78	- How long it takes vs how long the serial version would have taken
79	- The output files being created
80
81	If any worker fails or produces poor output, note which one and explain
82	how to rerun just that piece without redoing the whole batch.
83
84	### Step 5 — Teach the merging step
85
86	Parallel workers produce separate outputs. The final step is always a merge:
87
88	"Now tell Cowork: 'Read all the [output files] and combine them into a
89	single [format] saved as [final-filename].md.' This is the consolidation
90	step — it runs serially, but it only happens once, so it's fast."
91
92	Help the user write their specific merge prompt.
93
94	### Step 6 — Give them the parallel-ready checklist
95
96	Leave them with a mental checklist they can apply to future tasks:
97
98	1. Can this be split into N independent pieces? If yes, use parallel workers.
99	2. Does each piece produce a file as its output? If yes, add a merge step.
100	3. Are the pieces roughly the same size? If wildly different, split the
101	biggest piece further before running.
102	4. Is N greater than 5? Parallel workers save the most time on 5+ items.
103
104	## Quality Checks
105
106	Before finishing, verify:
107
108	- [ ] The user's task was confirmed as parallel-ready (independent pieces)
109	- [ ] A time comparison was given (serial vs parallel estimate)
110	- [ ] The request was rewritten using the p

People who liked this also viewed

6 picks

Type