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.
npx agentmods add agents/buttjer/n8n-decanter/offline-loopgit clone --depth 1 https://github.com/buttjer/n8n-decanterWhat 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 | $0.00015 | $0.01012 |
| Opus 5 | $0.00008 | $0.00506 |
| Sonnet 5 | $0.00003 | $0.00202 |
| Haiku 4.5 | $0.00002 | $0.00101 |
Grade A, and why
offline-loop scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured yesterday.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
What it actually says
Several verbs are fully offline — no credentials, no network, no live n8n — which makes them safe for agents to run without supervision:
preflight --offline— the static tier on its own: the layout-compliance guard (layout) plus the typecheck (types, the same wrapper that maps top-level-returnnode bodies back to real line numbers).--offlinedrops the instance tier, so nothing is read from n8n. Run it after editing any code file; treat anot readyverdict as a blocker. Every violation is listed in the failing check's indenteddetails— the one-line message is only the summary.node run— executes a node's body against a faked n8n context and prints the items it returns. With a fixture,$input,$('Node Name'), env, and static data are all controllable — real execution feedback without touching the instance.
Adding --simulate (preflight --offline --simulate) keeps the loop
credential-free and still never contacts n8n, but it boots a throwaway local
engine to really run the workflow — Docker, and minutes rather than
milliseconds. An occasional deeper pass, not the per-edit one.
The first check of a task is the one exception — run it online. A bare
preflight before your first edit reports whether
someone changed that code on the instance meanwhile (drift → pull and carry
on; CONFLICT → show the user diff first). --offline
cannot answer that — it drops the instance tier by definition — so the
credential-free loop below is what you run per edit, not what you open with.
A typical agent iteration:
# orient first: read-only, and the only check that sees the instance's side
n8n-decanter preflight order-sync
# after editing code/parse-order.ts and workflow.json
n8n-decanter node run workflows/order-sync/code/parse-order.ts fixture.json
n8n-decanter preflight --offline
# both green -> push: the draft is where the work lands, and code that only
# exists in this folder is not done
n8n-decanter push order-sync
# now the draft holds your code -> grade it on the instance. This step leaves
# the offline loop: `test` grades the DRAFT on n8n, so it only means anything
# once you have pushed (before a push it would grade the old code). Bare, it is
# a static check and executes nothing; add --scenario/--execution for a real run.
n8n-decanter test order-sync
# going LIVE (`publish` / `push --publish`) stays the user's call
Adding a Code node from scratch is a structure act — it happens in n8n
(the editor, or an addNode MCP op through the
guard with no jsCode), then
pull lands it as an empty code/ file with its
placeholder and state entry (the node lands disconnected; wire it in n8n).
Write the code in the file, verify with node run + preflight --offline,
and the first push seeds the node's source. The
sync layout page shows the shapes.
Because verification routes through the CLI, n8n-decanter must be on the
sync dir's PATH — see Installation.
Exit codes: one gate, one view
preflight is the gate — exit 1 when the verdict is not ready (any
check failed), 0 otherwise; --fail-on=warn makes a caution fail too.
That is the exit code to branch on, offline or online.
diff is the view — the per-node line diff between
your files and the n8n draft — and it always exits 0, like git diff.
Never read a clean diff exit as a passing check. (It is also not offline: it
reads the draft from the instance.)
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.
- yesterday First seen · 76 lines · 15 tokens per session scan A 2ad008b4450a
offline-loop is an agent published in the GitHub repository buttjer/n8n-decanter (2 stars, last pushed yesterday), licensed MIT. It adds 15 tokens to every session and 1,012 once invoked, about $0.0001 per session on Opus 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 agents, from other repositories
linear-issue-triager
Use this agent proactively when a Linear issue is created, updated, or needs comprehensive analysis. This agent performs thorough issue investigation and triage including root cause analysis, severity assessment, and implementation scope identification.
AGENTS
Conventions for the @n8n/agents package.
developer
Use this agent for any n8n development task - frontend (Vue 3), backend (Node.js/TypeScript), workflow engine, node creation, or full-stack features. The agent automatically applies n8n conventions and best practices. Examples: user: 'Add a new button to the workflow editor' assistant: 'I'll use the developer agent to…
form
FORM — Design. Visual system, UI architecture, motion, accessibility, design PRDs. Use when a human will see the work, when a landing page or product UI is requested, or when Forge is about to invent a look. Use proactively for any user-facing surface.
aria
ARIA — AI engineering. New agents, skills, evals, MCP wiring, department growth. Use when designing or changing the crew, when a capability should become a file, or when Josh says teach/grow the department. Use proactively when a one-off prompt is about to become a standing job.
axis
AXIS — Architecture. System design, ADRs, irreversible calls, architectural review. Use when the shape of the system is unset, when a shortcut would lock us in, or before Forge writes a feature that spans boundaries. Use proactively for any new product or service.