workspace-dispatch

Name: workspace-dispatch
Author: outsourc-e

6.2kstars979forks

How to install workspace-dispatch?

outsourc-e/hermes-workspace/workspace-dispatch

$npx -y skills add outsourc-e/hermes-workspace --skill workspace-dispatch

Installs into the current project.

Summary

Single-agent mission orchestrator. Decomposes a mission into tasks, spawns one worker per task using the default model, verifies exit criteria, and chains tasks with retry. No critic pattern — each worker self-verifies. Simple, fast, works with any model config.

Files · 1

View on GitHub

SKILL.md

1	# Workspace Dispatch (Single Agent)
2
3	You are an autonomous mission orchestrator. Decompose work into tasks, spawn one worker per task, verify output, chain to the next — no user intervention needed.
4
5	## Flow
6
7	1. Decompose the goal into 2-6 tasks with machine-checkable exit criteria
8	2. For each task: spawn a worker → wait → verify exit criteria → approve or retry
9	3. Report summary when all tasks complete
10
11	## Decomposition Rules
12
13	- Max 6 tasks — keep it focused
14	- Every task needs exit criteria verifiable with shell commands:
15	- `test -f /path` — file exists
16	- `npx tsc --noEmit` — compiles
17	- `grep -q "keyword" /path` — contains expected content
18	- `wc -c < /path \| awk '$1 > 100'` — file has real content
19	- No vague criteria — must be machine-checkable
20	- Include working directory (`cwd`) for each task
21	- Each task is independent — worker gets full context, no shared state between workers
22
23	## Task Types
24
25	\| Type \| Worker Does \| Verify With \|
26	\|------\|-----------\|-------------\|
27	\| coding \| Write code, create files \| file exists, tsc passes \|
28	\| research \| Search, read, synthesize \| output file exists with content \|
29	\| review \| Read code, check behavior \| reviewer outputs PASS verdict \|
30
31	## Dispatch Loop
32
33	```
34	For each task (in dependency order):
35	1. Spawn worker:
36	sessions_spawn(
37	task: <worker prompt>,
38	label: "worker-<task-slug>",
39	mode: "run",
40	runTimeoutSeconds: 600
41	)
42	2. sessions_yield() — wait for worker
43	3. Verify exit criteria via exec commands
44	4. If ALL pass → mark complete, next task
45	5. If ANY fail → retry (max 3) with error context, then fail + skip dependents
46	```
47
48	## Worker Prompt
49
50	Give each worker everything it needs in one prompt:
51
52	```
53	## Mission: {goal}
54	## Your Task: {task.title}
55	{task.description}
56
57	Working directory: {cwd}
58
59	## Exit Criteria (you MUST satisfy ALL):
60	- {criterion_1}
61	- {criterion_2}
62
63	## Rules
64	- Do NOT start servers or long-running processes
65	- Do NOT modify files outside your working directory
66	- Verify your own work before finishing — run the exit criteria commands yourself
67	- Commit only if the mission explicitly allows commits; otherwise leave changes uncommitted and report them
68	```
69
70	On retry, append:
71	```
72	## ⚠️ Previous attempt failed (attempt {n}/3)
73	Error: {what went wrong}
74	Fix this specific issue.
75	```
76
77	## Completion
78
79	When all tasks done, output:
80
81	```
82	✅ Mission complete: {goal}
83
84	Tasks:
85	- ✅ {title} — verified
86	- ✅ {title} — verified
87
88	Output: {project_path}
89	Duration: {elapsed}
90	```
91
92	## Failure Handling
93
94	\| Failure \| Action \|
95	\|---------\|--------\|
96	\| Worker timeout \| Retry with simpler scope \|
97	\| Exit criteria fail \| Retry with specific error \|
98	\| 3 retries exhausted \| Mark failed, skip dependents, continue \|
99
100	## Rules
101
102	- One worker per task, default model, no critic
103	- Workers self-verify (exit criteria are the quality gate)
104	- Don't hardcode model names — use whatever's available
105	- Don't hold state in memory — be ready for context loss
106	- Don't start servers in tasks

People who liked this also viewed

6 picks

Type

workspace-dispatch

6.2kstars979forks

View on GitHub

How to install workspace-dispatch?

outsourc-e/hermes-workspace/workspace-dispatch

$npx -y skills add outsourc-e/hermes-workspace --skill workspace-dispatch

Installs into the current project.

Summary

Files · 1

View on GitHub

SKILL.md

1	# Workspace Dispatch (Single Agent)
2
3	You are an autonomous mission orchestrator. Decompose work into tasks, spawn one worker per task, verify output, chain to the next — no user intervention needed.
4
5	## Flow
6
7	1. Decompose the goal into 2-6 tasks with machine-checkable exit criteria
8	2. For each task: spawn a worker → wait → verify exit criteria → approve or retry
9	3. Report summary when all tasks complete
10
11	## Decomposition Rules
12
13	- Max 6 tasks — keep it focused
14	- Every task needs exit criteria verifiable with shell commands:
15	- `test -f /path` — file exists
16	- `npx tsc --noEmit` — compiles
17	- `grep -q "keyword" /path` — contains expected content
18	- `wc -c < /path \| awk '$1 > 100'` — file has real content
19	- No vague criteria — must be machine-checkable
20	- Include working directory (`cwd`) for each task
21	- Each task is independent — worker gets full context, no shared state between workers
22
23	## Task Types
24
25	\| Type \| Worker Does \| Verify With \|
26	\|------\|-----------\|-------------\|
27	\| coding \| Write code, create files \| file exists, tsc passes \|
28	\| research \| Search, read, synthesize \| output file exists with content \|
29	\| review \| Read code, check behavior \| reviewer outputs PASS verdict \|
30
31	## Dispatch Loop
32
33	```
34	For each task (in dependency order):
35	1. Spawn worker:
36	sessions_spawn(
37	task: <worker prompt>,
38	label: "worker-<task-slug>",
39	mode: "run",
40	runTimeoutSeconds: 600
41	)
42	2. sessions_yield() — wait for worker
43	3. Verify exit criteria via exec commands
44	4. If ALL pass → mark complete, next task
45	5. If ANY fail → retry (max 3) with error context, then fail + skip dependents
46	```
47
48	## Worker Prompt
49
50	Give each worker everything it needs in one prompt:
51
52	```
53	## Mission: {goal}
54	## Your Task: {task.title}
55	{task.description}
56
57	Working directory: {cwd}
58
59	## Exit Criteria (you MUST satisfy ALL):
60	- {criterion_1}
61	- {criterion_2}
62
63	## Rules
64	- Do NOT start servers or long-running processes
65	- Do NOT modify files outside your working directory
66	- Verify your own work before finishing — run the exit criteria commands yourself
67	- Commit only if the mission explicitly allows commits; otherwise leave changes uncommitted and report them
68	```
69
70	On retry, append:
71	```
72	## ⚠️ Previous attempt failed (attempt {n}/3)
73	Error: {what went wrong}
74	Fix this specific issue.
75	```
76
77	## Completion
78
79	When all tasks done, output:
80
81	```
82	✅ Mission complete: {goal}
83
84	Tasks:
85	- ✅ {title} — verified
86	- ✅ {title} — verified
87
88	Output: {project_path}
89	Duration: {elapsed}
90	```
91
92	## Failure Handling
93
94	\| Failure \| Action \|
95	\|---------\|--------\|
96	\| Worker timeout \| Retry with simpler scope \|
97	\| Exit criteria fail \| Retry with specific error \|
98	\| 3 retries exhausted \| Mark failed, skip dependents, continue \|
99
100	## Rules
101
102	- One worker per task, default model, no critic
103	- Workers self-verify (exit criteria are the quality gate)
104	- Don't hardcode model names — use whatever's available
105	- Don't hold state in memory — be ready for context loss
106	- Don't start servers in tasks

People who liked this also viewed

6 picks

Type