execute-fleet

Name: execute-fleet
Author: anombyte93

582stars57forks

How to install execute-fleet?

anombyte93/prd-taskmaster/execute-fleet

$npx -y skills add anombyte93/prd-taskmaster --skill execute-fleet

Installs into the current project.

Summary

Phase execution skill for licensed Atlas Fleet runs. Use when HANDOFF has selected Atlas Fleet and the project should be executed across isolated launcher worktrees with inbox-based result collection, verified CDD cards, sequential integration merges, and one final PR.

Files · 1

View on GitHub

SKILL.md

1	# execute-fleet
2
3	Atlas Fleet is the premium parallel sibling of `execute-task`. It keeps the
4	same proof discipline, but the orchestrator owns the scoreboard while workers
5	only build inside isolated worktrees.
6
7	## Hard Gates
8
9	Before the first wave, all gates must pass. If any gate fails, report the gap
10	and stop; do not fall back to solo execution from inside this skill.
11
12	1. `mcp__plugin_prd_go__detect_capabilities()` reports
13	`tier: "premium"` and atlas-launcher MCP registration/aliveness.
14	2. `mcp__atlas-launcher__inbox_read` is callable for this session.
15	3. `.taskmaster/tasks/tasks.json` exists.
16	4. `.taskmaster/reports/task-complexity-report.json` exists.
17	5. `git status --short` is empty. Fleet starts only from a committed base.
18	6. The integration branch policy is clear: use `fleet-integration`; main is never auto-touched.
19
20	SOLE-WRITER RULE: only this orchestrator writes
21	`.taskmaster/tasks/tasks.json` and `.atlas-ai/state/pipeline.json`. Workers
22	must never edit those files. The orchestrator may update task state only
23	through TaskMaster or the plugin pipeline MCP, and only after verification.
24
25	## Wave Loop
26
27	Repeat until no runnable tasks remain:
28
29	1. Call `mcp__plugin_prd_go__compute_fleet_waves(concurrency=<N>, tag=<tag>)`.
30	Use the returned frontier as the only dispatch source. If it reports a
31	deadlock, render status, mark the blocked set, and stop dispatching those
32	tasks.
33	2. For each chunk in the current wave, spawn exactly one worker:
34	`mcp__atlas-launcher__session_spawn(isolation="worktree", report_to=<this session>, model=<routing[task_id] model part>, prompt=<worker prompt>)`.
35	Model is NEVER left default: `compute_fleet_waves` returns a `routing`
36	map (task id -> backend:model) from the capability ladder — Fable for the
37	hardest/longest-running (frontier tier), down the cost-efficiency curve to
38	haiku for trivial tasks. Pass the model part explicitly; non-claude backends
39	require experimental_backends=true and the launcher backend param.
40	The prompt must include the full task JSON inline; never tell workers to
41	read shared `tasks.json`.
42	3. Inspect the spawn result. If `prompt_injected` is false, re-kick once with
43	`mcp__atlas-launcher__session_send` using the same worker prompt, then
44	confirm injection/readiness. If it is still false, treat that worker as
45	failed and apply the retry policy below.
46	4. Render the fleet status view after the wave starts and after each wave
47	transition.
48	5. Poll `mcp__atlas-launcher__inbox_read` for terminal worker messages. The
49	only accepted terminal status vocabulary is `DONE`, `DONE_WITH_CONCERNS`,
50	`NEEDS_CONTEXT`, or `BLOCKED`; any other terminal word is a protocol
51	failure and counts as a worker failure.
52	6. On a completion message, verify the branch, never the narration:
53	- The worker branch contains `.atlas-ai/cdd/task-<id>.json`.
54	- No `.atlas-ai/evidence/` file in that branch contains a non-zero
55	`Exit status N` line.
56	- The branch changed only its own worktree scope and did not edit
57	`.taskmaster/tasks/tasks.json` or `.atlas-ai/state/pipeline.json`.
58	7. Never mark a task done without the CDD card. Missing card means the worker
59	did not satisfy the contract, regardless of any `DONE` message.
60	8. Merge verified worker branches into `fleet-integration` sequentially, one
61	at a time. After each merge, run the checker/build gate expected for the
62	project before merging the next branch.
63	9. Mark the task done only after the merge gate passes:
64	`python3 script.py set-status --id <id> --status done`.
65	10. Recompute waves after every accepted merge. Do not keep dispatching from a
66	stale frontier.
67
68	## Worker Prompt Template
69
70	Embed this template verbatim for each worker, replacing placeholders before
71	dispatch:
72
73	```text
74	WORKER_CONTRACT_ORCHESTRATOR_REPORT_TO
75	You are an Atlas Fleet worker. Your orchestrator is <ORCHESTRATOR_ID>. Report every question, blocker, and terminal result to <REPORT_TO_SESSION>.
76
77	WORKER_CONTRACT_FULL_TASK_JSON_INLINE
78	Your assigned task JSON is inline below. Treat this as the source of truth. Do not read shared .taskmaster/tasks/tasks.json.
79	<FULL_TASK_JSON>
80
81	WORKER_CONTRACT_WORKTREE_BRANCH
82	Work only in this isolated worktree and branch:
83	worktree: <WORKTREE_PATH>
84	branch: <WORKER_BRANCH>
85
86	WORKER_CONTRACT_CDD_CARD
87	Before reporting any terminal status, write this CDD card in your worktree: .atlas-ai/cdd/task-<id>.json. The card must list the checks you ran and the evidence paths that prove

People who liked this also viewed

6 picks

Type

execute-fleet

582stars57forks

View on GitHub

How to install execute-fleet?

anombyte93/prd-taskmaster/execute-fleet

$npx -y skills add anombyte93/prd-taskmaster --skill execute-fleet

Installs into the current project.

Summary

Files · 1

View on GitHub

SKILL.md

1	# execute-fleet
2
3	Atlas Fleet is the premium parallel sibling of `execute-task`. It keeps the
4	same proof discipline, but the orchestrator owns the scoreboard while workers
5	only build inside isolated worktrees.
6
7	## Hard Gates
8
9	Before the first wave, all gates must pass. If any gate fails, report the gap
10	and stop; do not fall back to solo execution from inside this skill.
11
12	1. `mcp__plugin_prd_go__detect_capabilities()` reports
13	`tier: "premium"` and atlas-launcher MCP registration/aliveness.
14	2. `mcp__atlas-launcher__inbox_read` is callable for this session.
15	3. `.taskmaster/tasks/tasks.json` exists.
16	4. `.taskmaster/reports/task-complexity-report.json` exists.
17	5. `git status --short` is empty. Fleet starts only from a committed base.
18	6. The integration branch policy is clear: use `fleet-integration`; main is never auto-touched.
19
20	SOLE-WRITER RULE: only this orchestrator writes
21	`.taskmaster/tasks/tasks.json` and `.atlas-ai/state/pipeline.json`. Workers
22	must never edit those files. The orchestrator may update task state only
23	through TaskMaster or the plugin pipeline MCP, and only after verification.
24
25	## Wave Loop
26
27	Repeat until no runnable tasks remain:
28
29	1. Call `mcp__plugin_prd_go__compute_fleet_waves(concurrency=<N>, tag=<tag>)`.
30	Use the returned frontier as the only dispatch source. If it reports a
31	deadlock, render status, mark the blocked set, and stop dispatching those
32	tasks.
33	2. For each chunk in the current wave, spawn exactly one worker:
34	`mcp__atlas-launcher__session_spawn(isolation="worktree", report_to=<this session>, model=<routing[task_id] model part>, prompt=<worker prompt>)`.
35	Model is NEVER left default: `compute_fleet_waves` returns a `routing`
36	map (task id -> backend:model) from the capability ladder — Fable for the
37	hardest/longest-running (frontier tier), down the cost-efficiency curve to
38	haiku for trivial tasks. Pass the model part explicitly; non-claude backends
39	require experimental_backends=true and the launcher backend param.
40	The prompt must include the full task JSON inline; never tell workers to
41	read shared `tasks.json`.
42	3. Inspect the spawn result. If `prompt_injected` is false, re-kick once with
43	`mcp__atlas-launcher__session_send` using the same worker prompt, then
44	confirm injection/readiness. If it is still false, treat that worker as
45	failed and apply the retry policy below.
46	4. Render the fleet status view after the wave starts and after each wave
47	transition.
48	5. Poll `mcp__atlas-launcher__inbox_read` for terminal worker messages. The
49	only accepted terminal status vocabulary is `DONE`, `DONE_WITH_CONCERNS`,
50	`NEEDS_CONTEXT`, or `BLOCKED`; any other terminal word is a protocol
51	failure and counts as a worker failure.
52	6. On a completion message, verify the branch, never the narration:
53	- The worker branch contains `.atlas-ai/cdd/task-<id>.json`.
54	- No `.atlas-ai/evidence/` file in that branch contains a non-zero
55	`Exit status N` line.
56	- The branch changed only its own worktree scope and did not edit
57	`.taskmaster/tasks/tasks.json` or `.atlas-ai/state/pipeline.json`.
58	7. Never mark a task done without the CDD card. Missing card means the worker
59	did not satisfy the contract, regardless of any `DONE` message.
60	8. Merge verified worker branches into `fleet-integration` sequentially, one
61	at a time. After each merge, run the checker/build gate expected for the
62	project before merging the next branch.
63	9. Mark the task done only after the merge gate passes:
64	`python3 script.py set-status --id <id> --status done`.
65	10. Recompute waves after every accepted merge. Do not keep dispatching from a
66	stale frontier.
67
68	## Worker Prompt Template
69
70	Embed this template verbatim for each worker, replacing placeholders before
71	dispatch:
72
73	```text
74	WORKER_CONTRACT_ORCHESTRATOR_REPORT_TO
75	You are an Atlas Fleet worker. Your orchestrator is <ORCHESTRATOR_ID>. Report every question, blocker, and terminal result to <REPORT_TO_SESSION>.
76
77	WORKER_CONTRACT_FULL_TASK_JSON_INLINE
78	Your assigned task JSON is inline below. Treat this as the source of truth. Do not read shared .taskmaster/tasks/tasks.json.
79	<FULL_TASK_JSON>
80
81	WORKER_CONTRACT_WORKTREE_BRANCH
82	Work only in this isolated worktree and branch:
83	worktree: <WORKTREE_PATH>
84	branch: <WORKER_BRANCH>
85
86	WORKER_CONTRACT_CDD_CARD
87	Before reporting any terminal status, write this CDD card in your worktree: .atlas-ai/cdd/task-<id>.json. The card must list the checks you ran and the evidence paths that prove

People who liked this also viewed

6 picks

Type