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 instructions/clidey/deptrust/agents-mdgit clone --depth 1 https://github.com/clidey/deptrustWhat 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 | $0.01305 | $0.01305 |
| Opus 5 | $0.00652 | $0.00652 |
| Sonnet 5 | $0.00261 | $0.00261 |
| Haiku 4.5 | $0.00130 | $0.00130 |
Grade A, and why
deptrust AGENTS.md 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 2d 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 — 43 lines — stays where its author put it; the contents beside it link to each section on GitHub.
What this is
deptrust is a Go CLI and stdio MCP server that checks package versions for known vulnerabilities across 13 ecosystems (npm, pypi, cargo, go, rubygems, nuget, maven, packagist, pub, cocoapods, hex, hackage, github-actions). It calls public registry and advisory APIs (OSV, GitHub Advisory DB) directly — there is no hosted backend. It also ships as an npm package (@clidey/deptrust) whose bin is a Node installer that downloads the Go binary and registers MCP/skill integrations.
Commands
go build -o deptrust ./cmd/deptrust # build the binary
go test ./... # all tests — NOTE: integration_test.go hits live registries/network
go test ./internal/risk/... # one package
go test ./internal/app/ -run TestName # one test
./lint.sh # golangci-lint with repo config (requires golangci-lint v2.12.2)
gofmt -l . # CI fails if this prints any file
CI (.github/workflows/ci.yml) runs gofmt -l, go test ./..., go build, and npm pack --dry-run. There is no -short gating — internal/registry/integration_test.go makes real network calls and runs as part of go test ./....
Architecture
The flow is CLI/MCP entrypoint → app.App orchestrator → registry resolver + vulnerability providers → risk scorer.
cmd/deptrust/main.go— CLI argument parsing only. Subcommands:check,suggest,compare,mcp,version. Maps errors to exit codes by string-matching the message (provider→2,not found→3,block→10).internal/app/app.go— the orchestrator (App).CheckPackageresolves the version via the registry, fans out to all vulnerability providers in parallel (goroutines + channel), dedupes advisories by GHSA/CVE id, computes non-CVE signals (recent-release, mutable GitHub Actions tags), then callsrisk.Score.SuggestSafeVersionandCompareVersionsbuild oncheckResolved.Appholds an injectablenow func() time.Timefor deterministic signal tests.internal/registry/— one file per ecosystem (npm.go,pypi.go, …), all dispatched fromregistry.go'sResolveswitch. Each returns aVersionInfo(resolved version, latest, full version list, publish timestamps).models.LatestVersionis the sentinel for "latest". Shared helpers incommon.go/http.go/errors.go.internal/osv/andinternal/github/— vulnerability providers. Both implement thevulnerabilityClientinterface (Name(),Query()) defined inapp.go. Providers may also implementecosystemAwareProvider(Supports(ecosystem)); unsupported ecosystems are reported asskipped_providersand driveadvisory_coverage(full/partial/none/error). When no provider covers an ecosystem, the result isunknown— never treated as safe.internal/risk/risk.go— pure scoring. Maps highest severity to a recommendation: critical/high→block, medium/unknown→review, low/none→allow. Recommendation constants live here and are referenced across packages.internal/models/models.go— shared types, theEcosystemenum,NormalizeEcosystem(string→enum), andOSVEcosystem(enum→OSV name).internal/mcp/server.go— hand-rolled JSON-RPC stdio MCP server (no SDK). Exposescheck_package,suggest_safe_version,compare_versions.compact.goproduces a slimmed-down response (counts + highest severity, no advisory bodies) to keep agent context small; the full response is available via thefull_response_commandfield.serverInstructionsis the prompt sent to agents oninitialize— keep it in sync with the README's MCP section.
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.
- 2d ago First seen · 43 lines · 1,305 tokens per session scan A eaab656ce1dc
deptrust AGENTS.md is an instructions file published in the GitHub repository clidey/deptrust (61 stars, last pushed 13d ago), licensed MIT. It adds 1,305 tokens to every session, about $0.0065 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 2026-08-30.
Other instructions, from other repositories
Sighthound AGENTS.md
Instructions for Corgea/Sighthound, covering claude, commands, python / django test naming and agent skill.
mcp-for-beginners AGENTS.md
Instructions for microsoft/mcp-for-beginners, covering agents.md, project overview, key technologies, architecture and setup commands.
conductor AGENTS.md
AGENTS.md instructions for conductor-oss/conductor, covering agents.md, project overview, setup commands, java version references and code style.
conductor CLAUDE.md
Claude Code instructions for conductor-oss/conductor, covering claude.md — conductor (server repo), writing documentation, for each content type, start here, when you can't verify and key source locations.
api2mcp4j CLAUDE.md
Claude Code instructions for TheEterna/api2mcp4j, covering claude.md, 项目概述, 顶层授权框架(必读), 开发心法与规范导航 and 通用心法(docs/rules/global/,所有会话默认生效).
lsp-tools copilot-instructions.md
Instructions for zircote/lsp-tools, covering copilot instructions, priorities, key files, commands and when adding language support.