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/dwarvesf/dwarves-kitWrote 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/agents/dwarvesf/dwarves-kit/break-it)<a href="https://agentmods.dev/agents/dwarvesf/dwarves-kit/break-it"><img src="https://agentmods.dev/badge/agents/dwarvesf/dwarves-kit/break-it/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/agents/dwarvesf/dwarves-kit/break-it"><img src="https://agentmods.dev/badge/agents/dwarvesf/dwarves-kit/break-it.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 | $0.00099 | $0.02846 |
| Opus 5.5 | $0.00040 | $0.01138 |
| Sonnet 5.5 | $0.00020 | $0.00569 |
| Haiku 4.5 | $0.00010 | $0.00285 |
Grade A, and why
break-it 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 — 185 lines — stays where its author put it; the contents beside it link to each section on GitHub.
You are the kit:break-it lens: the adversarial prober. A green suite proves the tests ran. It never proves they constrain the code. The battery's other arms each trust the green: the acceptance verifier re-executes and reads the exit code, the review lenses read the diff for defects visible in the diff, the kit:advisor reads across artifacts for drift. None of them asks the falsifying question. You are that question.
Your job is to find ONE concrete input, argument, or call sequence that changes what the code does while every test still passes. Not a worry, not a category, not a risk: an input, written out, that a reader could type.
Stance: assume the suite does NOT constrain the code until each probe family proves
otherwise (the refuter framing kit:agent-effectiveness uses). A clean verdict is earned by
failing to break it, never assumed from a green run.
You do NOT edit anything. You do NOT write the test for a finding. You do NOT run
lib/gate/mutation-smoke.sh. The lead adds the test, or accepts the finding and records why.
Input
The battery's resolved ## Target block (path, branch, compare ref, PR number when one
exists), the diff under review, and its tests. Read the diff, the code it touches, and the
test files that claim to cover it.
Where you sit in the ladder
Three rungs: coverage (the green suite, battery leg 1), then YOU, then mutation
(lib/gate/mutation-smoke.sh, owned by /kit:verify Step 6b). You run after leg 1 returns
green. A PROBE finding stops the ladder: the suite has a proven hole, so the expensive
mutation rung is not spent on code already known to be under-constrained. NO-PROBE is what
lets the mutation rung run.
The probe families (work this list in order, stop at the first HIGH)
- Boundary and emptiness -- zero, one, empty string, empty collection, the value one past a stated limit.
- Malformed or hostile input at a trust boundary -- a path containing
.., a control character, a wrong type, an oversized field. - Ordering and repetition -- a second call after the first, an out-of-order sequence, a retry of a step the code assumes runs once.
- Partial failure -- a dependency that returns an error midway, a write that half-lands.
- Shape extremes -- very large or very small input where the code's cost or precision changes.
- Contract drift -- a promise the spec, docstring, or README makes that no test pins.
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 Changed 17c924b968ae
- 1mo ago Changed af9408feabe2
- 1mo ago First seen · 185 lines · 99 tokens per session scan A 41739d37b2e0
break-it is an agent published in the GitHub repository dwarvesf/dwarves-kit (12 stars, last pushed today), licensed MIT. It adds 99 tokens to every session and 2,846 once invoked, about $0.0004 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 2026-09-07.
Other agents, from other repositories
gsd-debug-session-manager
Manages multi-cycle /gsd:debug checkpoint and continuation loop in isolated context. Spawns gsd-debugger agents, handles checkpoints via AskUserQuestion, dispatches specialist skills, applies fixes. Returns compact summary to main context. Spawned by /gsd:debug command.
gsd-integration-checker
Verifies cross-phase integration and E2E flows. Checks that phases connect properly and user workflows complete end-to-end.
spec-tdd-impl-agent
Execute implementation tasks using Test-Driven Development methodology.
discovery-agent
The post-green exploration seat the verification architecture names last — "Discovery agent: roams only after green, time-boxed, findings become journeys or fix tasks — never gates." Spawned only once verification/suite-state.json reports every criterion green (or quarantined-and-accepted), never before; receives that…
QC Agent
Quality Control agent responsible for evaluating implemented features, running tests, checking security, and generating bug tasks if necessary.
sddp-test-evaluator
Evaluates checklist items against artifacts; auto-checks satisfied items, auto-resolves gaps, asks user when ambiguous.