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 skills/megaprompting/torque-loop/loopnpx skills add Megaprompting/torque-loop --skill loopgit clone --depth 1 https://github.com/Megaprompting/torque-loopWhat 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 | $0.00076 | $0.00748 |
| Opus 5 | $0.00038 | $0.00374 |
| Sonnet 5 | $0.00015 | $0.00150 |
| Haiku 4.5 | $0.00008 | $0.00075 |
Grade A, and why
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.
How it starts
The opening of the file, as written. The whole thing — 75 lines — stays where its author put it; the contents beside it link to each section on GitHub.
/ratchet:loop — recursive advancement
One pass rarely finishes anything. This command repeats the core cycle until the artifact actually holds under attack. It is the closest thing Ratchet has to a software-quality agent: discover the missing thing, build it, try to break it, fix what broke, serialize.
Step 0 — Load state
ratchet status
ratchet score confidence
The cycle (repeat until the stop condition holds)
Each iteration:
- Discover the missing thing. The highest-leverage gap right now — an untested
assumption, an open defect, a missing piece of the artifact. (Borrow
/ratchet:cutand/ratchet:auctionlogic to pick it.) - Build the artifact or the fix (
/ratchet:build). - Attack it with the five-voice board (
/ratchet:attack); record defects. - Patch only what failed (
/ratchet:patch); resolve the original defect with evidence. - Verify bound to the artifact (
/ratchet:verify); on green,ratchet artifact close <id>. - Compile the advance (
/ratchet:compile); recompute confidence.
The cycle is build → attack → patch → verify → compile. Compiling before verifying records an unproven state as if it were an advance.
Report each iteration compactly:
ITERATION n: discovered <gap> → built <thing> → attack <c crit / h high> → patched <delta> → confidence <score>
Stop condition (all must hold)
Run ratchet score confidence and read BOTH loopClear and the workflow-closure layer
(closure.closed). Stop only when:
loopClearholds — no unresolved critical/high defect, no untested core assumption, no missing next action.workflowClosed.closedis true — the active artifact carries a closure certificate and no open defect is unattached.- The artifact is genuinely usable (passes its 5-point test).
If either read is false, run another iteration. Do not stop because the output "looks good" — the CLI reads are the arbiter, not your impression.
There is no convergence escape. If two iterations find nothing new while
workflowClosed.closed is false, you are not converged — you are blocked. Stop with
STOP REASON: blocked-needs-human, name the blocker the closure read printed, and hand it
over. "Nothing left to try" and "finished" are different states, and only one of them is
an ending.
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 · 75 lines · 76 tokens per session scan A b47974f877cc
loop is a skill published in the GitHub repository Megaprompting/torque-loop (5 stars, last pushed 1mo ago), licensed MIT. It adds 76 tokens to every session and 748 once invoked, about $0.0004 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 skills, from other repositories
new-plugin
Factory line for adding a new HAR verification plugin (like playwright or rocketsim) for any framework — research the framework docs, build the template under src/templates/plugins/, register it everywhere, validate on a real repository, and open a PR. Use when asked to add/create a plugin, plugin template, or…
factory-line
Factory line for executing one station of a declared multi-station program — read the installed line bundle (har line status), plan parallel work into isolated HAR slots, run the cumulative gate with har line gate, and hand off for human review. Use when asked to "run a factory line", "run the next station", "execute…
v1-milestone
Factory line for executing one milestone of the HAR v1.0.0 refactor (epic os-factory/har#225) — plan the wave of parallel subagents, implement each issue in its own HAR slot, ship stacked PRs, run the fixture-e2e milestone gate, and hand off for review. Use when asked to "run the next v1 milestone", "work on v1.0.0"…
ctx
Codebase intelligence and evidence-driven governance with the indexed ctx CLI. Use when exploring an unfamiliar repository, locating symbols or callers, checking for existing implementations, estimating change impact, enforcing architecture rules, scoring a branch, finding hotspots or duplication, or analyzing…
ctx
Codebase intelligence and evidence-driven governance with the indexed ctx CLI. Use when exploring an unfamiliar repository, locating symbols or callers, checking for existing implementations, estimating change impact, enforcing architecture rules, scoring a branch, finding hotspots or duplication, or analyzing…
golden-rss
Use when testing the rss golden build.