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/arturogarrido/claudinhoWrote 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/rules/arturogarrido/claudinho/trust-boundary)<a href="https://agentmods.dev/rules/arturogarrido/claudinho/trust-boundary"><img src="https://agentmods.dev/badge/rules/arturogarrido/claudinho/trust-boundary.svg" alt="Measured on agentmods" 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 | $0.00000 | $0.01040 |
| Opus 5 | $0.00000 | $0.00520 |
| Sonnet 5 | $0.00000 | $0.00208 |
| Haiku 4.5 | $0.00000 | $0.00104 |
Grade A, and why
trust-boundary 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 8d 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 — 70 lines — stays where its author put it; the contents beside it link to each section on GitHub.
The trust boundary
Feed responses and local cache files are untrusted input. They enter through
packages/core/src/trust/ and leave as a domain type or as a stated reason they
could not. Nothing else builds a Match or a MarketSignal from raw input.
Rules
- An adapter FETCHES; it does not interpret. Parsing lives in
trust/(trust/espn.tsis the model). The adapter that also parsed is howidandkickoffended up copied verbatim while every string beside them was cleaned. - Record refusal is local, not a provider outage.
ProviderAdapterreturns plain arrays. Omit malformed, duplicate, or truncated records while keeping readable siblings usable and attributed to the provider. Throw into the degraded path only for transport/JSON failure, or an incomplete payload with no usable records. Parser-localBoundedList.completeis diagnostic; never promote one refused record into a batch-wide blackout. - Both paths end at the same constructor. Live and cache reads call
sealMatch/sealMarketSignal. When they had separate rules, each fix landed on one of them;core/test/trust-parity.test.tsasserts they agree, as a JSON round trip, because that is literally what the cache file is. - Text has roles.
humanLabelis prose — no controls, no format characters, no emoji, bounded by display columns and code points.opaqueIdis checked against an exact grammar.canonicalTimestampre-emits one form and refuses a date that does not exist.productFlagGENERATES the flag from the nation. - Never accept a product glyph from input. While flags travelled through the
text filter it needed an emoji carve-out, and a carve-out without a grammar is
a covert channel: TAG characters, variation selectors and ZWJ each rode through
it. A
🏴plus 42 tag characters is ONE two-column glyph spelling a full instruction sentence — invisible on a terminal, legible to a model reading--json. - Say which kind of failure it is.
ParseResult=valid/definitive-none/malformed/ambiguous/unresolved. For per-item market resolution,valid,definitive-noneandambiguousare stable and cacheable;malformedandunresolvedmust not become definitive negatives. A successful provider batch with one malformed sibling is still shared for the coalescing TTL: identical bytes will parse identically, and immediate refetches only waste egress and starve the readable records. - Enrichment completeness reaches the renderer. Market cache/fetch helpers
return
{ signals, complete }; default-on annotations, dedicated tools, share cards, and structured output retain it. Only a complete empty batch may say "no signal"; an incomplete one warns and carriesmarketComplete:falseorcomplete:false. - Bound the work, not just the output.
takeBoundedslices before the per-record map. A surface reportstotal/shown/truncatedfrom oneBoundedList; no handler recomputes a count.totalis exact only whencompleteis true, so incomplete hot-path scans use a nonnumericmoremarker rather than guessing how many valid records remain. MCP response limiting preflights depth, width, entries, containers, and aggregate text before serialization or recursive shrinking; a byte cap alone does not bound the CPU and allocation spent discovering that a payload is too large. - Fail closed on ABSENCE. A missing field must be at least as rejecting as a
wrong one. Five market gates were
x != null && ..., so the more malformed payload was the more successful one.
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.
- 8d ago First seen · 70 lines · 0 tokens per session scan A dfc6961a3880
trust-boundary is a cursor rule published in the GitHub repository arturogarrido/claudinho (28 stars, last pushed yesterday), licensed MIT. It costs nothing until one of its globs matches a file; then it loads 1,040 tokens. 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 cursor rules, from other repositories
proxy-auth-transport
Every CLI command or browser UI action that calls a crit-web API must work in both modes.
lap
LAP CLI -- search, download, and install API specs for AI agents. Use when user needs an API (Stripe, Plaid, Twilio, etc). Commands: (1) npx -y @lap-platform/lapsh search , (2) npx -y @lap-platform/lapsh get -o .lap, (3) npx -y @lap-platform/lapsh skill-install --target cursor. The .lap file IS the spec -- do NOT…
gateway
Messenger gateway architecture, session store, and Telegram adapter conventions.
cursor
You are working on the checkout service. Preserve transaction integrity and auditability.
hatch3r-resilience-patterns
Resilience patterns in user code — circuit breakers, retry with decorrelated jitter, timeouts with deadline propagation, idempotency keys, bulkheads, hedged requests.
dev
AIOX agent @dev - Full Stack Developer.