Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
git clone --depth 1 https://github.com/damusix/atomic-claudeWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/commands/damusix/atomic-claude/atomic-plan)<a href="https://agentmods.dev/commands/damusix/atomic-claude/atomic-plan"><img src="https://agentmods.dev/badge/commands/damusix/atomic-claude/atomic-plan/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/commands/damusix/atomic-claude/atomic-plan"><img src="https://agentmods.dev/badge/commands/damusix/atomic-claude/atomic-plan.svg" alt="Reviewed on agentmods" width="80" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00054 | $0.05097 |
| Opus 5.5 | $0.00022 | $0.02039 |
| Sonnet 5.5 | $0.00011 | $0.01019 |
| Haiku 4.5 | $0.00005 | $0.00510 |
Grade A, and why
atomic-plan scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 1mo ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 264 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Plan before coding. Output is a spec — a checkpoint-table contract, not prose. Lives in docs/ because it's human-facing — .claude/.scratchpad/ is for LLM working memory, not user-invoked plans.
/atomic-plan is one command with one internal phase boundary. Lightweight tasks get an inline spec. Non-trivial tasks get a full design + spec-authoring loop with subagents. The LLM gauges triviality; the user confirms when uncertain.
Phases
Classify triviality
Gauge the task against this tier table. High confidence required for trivial — any uncertain signal pushes to borderline.
| Tier | Signals | Flow |
|---|---|---|
| Trivial | One cohesive slice; ≤3 files; no architectural choice; no new package/module; no cross-cutting concern; no public-contract change; one obvious approach; no business-logic shape to think through. | Inline spec only. No design doc. No subagent loop. May also offer: "small enough to skip planning — implement directly?" |
| Borderline | 4-8 files; one subsystem but unfamiliar code; small architectural or business-rule choice; partial unknowns; one viable approach with caveats. | Ask the user. Surface the signals + your lean; let them pick full / inline / abort. |
| Non-trivial (default) | Cross-system; new package/module; multiple viable approaches; touches existing contracts; phrased as "design X" / "rebuild Y" / "refactor Z"; >8 files; doesn't fit one cohesion slice; non-obvious business rules or feature shape to work out. | Full flow: Ground → Diverge → Design → optional pressure-test → Spec loop. |
Borderline confirmation format — one block, not a question series:
This looks borderline. Signals I'm weighing:
- <N> files across <areas>
- <architectural decision or unknown>
- <familiarity / scope notes>
Lean: <full | inline>.
Override? [full | inline | abort]
Trivial confirmation format — one line, silent confidence, no question mark:
Trivial — <one-line shape>. Writing spec inline, skipping loop.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 1mo ago First seen · 264 lines · 54 tokens per session scan A baa36aea9113
atomic-plan is a command published in the GitHub repository damusix/atomic-claude (113 stars, last pushed today), licensed MIT. It adds 54 tokens to every session and 5,097 once invoked, about $0.0002 per session on Opus 5.5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other commands, from other repositories
save
Save current work state for next session. Creates/updates .planning/ files (CHECKPOINT.md, STATE.md, settings.json) so Heimdall resumes with full context. NOT a rewind — saves forward progress. Run before closing a session or at any milestone.
designmatch
Bootstrap the designmatch skill in a React Native project from a Claude Design link. Auto-imports the Visual QA stub helper (visual-qa.ts), writes default state.vqa.json, registers the canonical source, updates .gitignore, and prints the wiring snippet. Use when starting visual-parity work on a new RN screen or when…
maintain
Run automated maintenance — seeker finds bugs from pod logs and raises GitHub issues, fixer picks them up and creates PRs. Can run as a one-shot or scheduled via /schedule.
bench
Reproduce the public benchmark table on your own machine (spec S-5). Runs a fixed suite of coding tasks under two arms — raw Claude Code vs Heimdall — and PRODUCES real numbers (tokens, wall time, tests passing, cost) by actually running them. Never hardcodes results. Safe by default (dry mode: validates the suite +…
autonomy
Set Heimdall autonomy (1=Guided, 2=Checkpoint, 3=Full Auto) — how much the agent does before asking. Use with a number, +/- to cycle, or no argument to show current.
level
Deprecated alias for /hmd:autonomy. Sets Heimdall autonomy (1=Guided, 2=Checkpoint, 3=Full Auto). Still works; prints a one-line rename notice.