atomic-planner
Skill0 starsChanged 4 months agoMITView on GitHub
Break a workstream or feature into atomic tasks (1-3 hours each, single-purpose, testable). Writes a markdown plan AND creates a native Task List for compaction-survival. Use when planning a workstream, decomposing a complex effort, or when the user asks "break this down" or "what are the steps".
In this repository5 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- freshenSkill
- session-closerSkill
- session-startSkill
The file
We split this skill into the rules it gives an agent
1 rule. About 0 are common, 1 is distinctive, and 0 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈1
Distinctive · estimate
This author's own choices. Worth a look.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 29
Create a native Task List in parallel: After writing the markdown plan, use TaskCreate for each AT-NNN.
≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
atomic-planner
When to use
- A task is too large to execute in one session.
- Multiple independent pieces of work need parallel handling.
- A complex synthesis needs sub-tasks.
- The user asks "break this down" or "what are the steps".
Process
- Read context —
docs/plan.md(the plan of record, v6),memory/NEXT_SESSION.md(current handoff),WORK_PROGRESS.md, and any in-flight task. For data-lookup / receipt / agent work, readdata/README.mdand the affectedsrc/arbiter/*.py. - Decompose into atomic tasks. Each task must be:
- Atomic — single concern, completable in 1-3 hours.
- Testable — has a "done when" criterion.
- Independent — minimal coupling to other in-flight tasks.
- Concrete — not "research X", but "produce file Z" or "run command Q".
- Honest about runtime-error causes — if a card names the root cause of a runtime error (a traceback, crash, or wrong output) that you inferred from a static read, tag it "hypothesis — reproduce before fixing", not as fact. The bug's true location only reveals itself at runtime; the implementing session must falsify the hypothesis cheaply (probe/traceback) before the first edit.
- Write task cards to
docs/plans/<topic>_atomic.mdusingreferences/task-template.md. - Create a native Task List in parallel. After writing the markdown plan, use TaskCreate for each AT-NNN. Dual representation is the point.
- Map dependencies. Annotate parallel-eligible tasks. Default to parallel-eligible when two tasks don't share inputs, outputs, or side effects.
- Report: total tasks, parallel count, recommended first task, estimated total time.
Anti-patterns (reject during decomposition)
- ❌ Tasks longer than 3 hours — break further.
- ❌ Tasks without a "done when" — not atomic.
- ❌ Vague verbs ("research", "look into", "explore") — replace with deliverables.
- ❌ Re-creating tasks already enumerated in
docs/plan.mdor an existingdocs/plans/*_atomic.md— extend the existing card instead. - ❌ Markdown plan without a native Task List — breaks under compaction.
- ❌ Stating a statically-inferred root cause of a runtime error as fact — tag it "hypothesis — reproduce before fixing".
Output
docs/plans/<topic>_atomic.md(markdown plan, human-readable)- Native Task List (runtime state)
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

