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/naimkatiman/continuous-improvementWrote 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/naimkatiman/continuous-improvement/swarm)<a href="https://agentmods.dev/commands/naimkatiman/continuous-improvement/swarm"><img src="https://agentmods.dev/badge/commands/naimkatiman/continuous-improvement/swarm/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/naimkatiman/continuous-improvement/swarm"><img src="https://agentmods.dev/badge/commands/naimkatiman/continuous-improvement/swarm.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 | $0.00052 | $0.01432 |
| Opus 5.5 | $0.00021 | $0.00573 |
| Sonnet 5.5 | $0.00010 | $0.00286 |
| Haiku 4.5 | $0.00005 | $0.00143 |
Grade A, and why
swarm 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 — 102 lines — stays where its author put it; the contents beside it link to each section on GitHub.
/swarm
Parallel-agent fan-out for evidence-based decision-making. Driven by the user's session report: provider-swap epics (Trading Economics → Forex Factory, ETF → CFD, Finnhub → R StocksTrader, Stooq removal) each took multiple sessions and hit fallback bugs in production because alternatives were not vetted upfront.
This command does not replace the superpowers:dispatching-parallel-agents skill — it builds on top of it with a contract-pinned, worktree-isolated, comparison-report shape.
When to use
The right trigger is "I need to pick between N options and I have a contract test that defines what 'works' means." Examples:
- Evaluating N candidate providers behind a shared interface (the report's recurring need)
- Trying N candidate refactors against the same test suite
- Prototyping N candidate library upgrades and measuring which breaks the fewest tests
- Generating N candidate UI designs that all pass the same accessibility + performance budget
Do NOT use this command for:
- Open-ended exploration without a contract — use
superpowers:brainstorminginstead - Single-implementation tasks — use the
superpowers:dispatching-parallel-agentsskill directly - Tasks where the candidates have different contracts — that is N separate jobs, not a swarm
Preconditions
- A shared contract test exists at a path the agents can reference (e.g.
tests/contracts/<name>.test.ts). - Candidates are listed up front (3-6 typical; 8 max). Letting agents propose candidates is allowed but the list freezes before fan-out.
- The base branch is clean.
- Each candidate has a clear evaluation rubric: PASS/FAIL per contract test, plus per-candidate measurements (latency, coverage, cost, free-tier limits, etc. — domain-specific).
Behavior
- Plan — restate the objective, list candidates, confirm contract test path. If unclear, halt and ask.
- Bootstrap — for each candidate, create an isolated worktree off
origin/main:git worktree add -b swarm/<objective>-<candidate> ../<candidate>. Pin the base SHA in the swarm log. - Fan out — dispatch one fresh sub-agent per candidate via the
Agenttool, in a single message (parallel, not serial). Each agent receives the contract test path, the candidate name, the rubric, and the worktree path. - Run — each agent implements the candidate behind the shared interface, runs the contract test against it, captures measurements per the rubric, and writes findings to
reports/<objective>/<candidate>.md. - Synthesize — when all agents complete, this command produces a decision matrix (candidates × rubric metrics) and recommends a winner with citations to the per-candidate reports.
- Stop — does NOT merge any candidate's worktree. Does NOT modify production. Output is evidence; the operator decides which candidate to advance.
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 · 102 lines · 52 tokens per session scan A c93ded2dada9
swarm is a command published in the GitHub repository naimkatiman/continuous-improvement (9 stars, last pushed 1mo ago), licensed MIT. It adds 52 tokens to every session and 1,432 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 2026-08-31.
Other commands, from other repositories
ap-swarm
Run a MultiAgentRuntime swarm in file mode — local journal at .runtime/conversation.md, no network.
reflect-promote
Promote learnings from skills to global CLAUDE.md.
reflect-meta
Show meta-learning statistics and pattern performance.
reflect-stats
Show cross-skill learning statistics.
checklist
Generate a custom checklist for the current feature based on user requirements.
clarify
Identify underspecified areas in the current feature spec by asking up to 5 highly targeted clarification questions and encoding answers back into the spec.