dsh-industry-research
AGENTS.md220 starsChanged 3 months agoApache-2.0View on GitHub
Standalone DeepSeek Harness plugin repository (dsh-industry-research). Development follows the dsh-plugin-guide skill and the official plugin contract; this file records repo-local decisions.
Where files are headingNew AGENTS.md files have more rules about keeping secrets out of the code, and fewer about code style.
Of the AGENTS.md files we read, the 31,439 first seen in the last six months, against the older ones: the kind of rule that gained most share, and the kind that lost most. About files like this in general, not this one, and an estimate, since a kind is a pattern in the words.
In this repository2 other files it gives its agents.
The file
We split this AGENTS.md into the rules it gives an agent
3 rules. About 0 are common, 0 are distinctive, and 3 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈0
Distinctive · estimate
This author's own choices. Worth a look.
3
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 29
No network outside
ctx.web: All retrieval goes through the official seam (provider selection, timeouts, and error taxonomy have one … - Very specific: line 32
No tunables hardcoded: Every knob is a validated Config field with a default in src/config.ts, an inline comment in cordis.patch.yml, and a …
- Very specific: line 34
This plugin registers no waterfall listeners: If one is ever added, allow/passthrough MUST call next().
≈ “Common” is an estimate: at least 1 in 5 of the 40,246 AGENTS.md files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
AGENTS.md
Standalone DeepSeek Harness plugin repository (dsh-industry-research). Development follows the dsh-plugin-guide skill and the official plugin contract; this file records repo-local decisions.
Layout
src/index.ts— function-plugin contract (name/inject/Config/apply; NO default export). Injectsskillsandtoolsonly. Publishes the packagedskills/directory through the officialFileSystemSkillProviderand registers the four tools.src/config.ts— Schemastery schema + explicitresolveConfig(no hidden?? defaultin run paths). Bounds fail loud at mount.src/chain.ts— theChainMapdata model + validation (pure): unique node ids, legal tiers, no dangling edges, and everyvaluerequires asourceRef(a value-less slot is an explicit gap).src/timeline.ts— thetimeline.jsonlstore (pure fs): URL-normalized dedupe, host allow/block lists, corrupt-line tolerance with a surfaced count, and the retention cap.src/sources.ts— the per-industrysources.jsonregistry: stableS<n>refs, origins, and SHA-256 content hashes.src/company.ts— company scan: text-format data files (no PDF), Markdown outlines, figure-candidate lines,card.json+card.md.src/report.ts— report assembly: evidence/sections/claims building, draft validation, the mechanical auto-draft, and the builtin-fallback Markdown + manifest renderer.src/engine-bridge.ts— the optionalctx.researchReportcontract, imported as types only fromdsh-research-report(erased at build time — the built entry never imports it at runtime) plus the structuralctx.get('researchReport')lookup behind anisResearchReportLikeguard. Never injected, never reached through a runtime import.src/web.ts— the structuralctx.weblookup, the offline/unmounted loud failures, and per-request timeout signals.src/paths.ts/src/toolkit.ts— workspace containment and the<cwd>/<industryRoot>/layout.src/events.ts— typed Cordis events (industry-research/map|track|report,@mode emit).src/tools/*.ts— the fourdefineTooldefinitions.skills/— the two methodology SKILL.md bundles (Chinese edition).fixtures/baijiu/— committed teaching fixtures for the keyless e2e suite (clearly fictional).scripts/—prepare.mjs(build),verify-self-contained.mjs,verify-artifacts.mjs,check-readme-sync.mjs(five-language gate),check-skills.mjs(skills frontmatter gate),fix-dts.mjs(declaration-specifier rewrite),changelog-section.mjs,loader-runner.mjs(real Loader composition runner).test/— vitest; REALContext/SessionStore/Session/ToolRuntime/SkillRegistry/WebRuntimefrom the 0.1.7-rc.2 peers. Only the pluggable edges (web providers, the optional report engine) are scripted, through the real registration mechanisms.
Hard rules applied here
- Optional capabilities are looked up, never injected.
ctx.webandctx.researchReportare resolved withctx.get(...)at execution time; without themindustry_trackfails loud with mount guidance andindustry_reporttakes the honest builtin-fallback path. Writing them intoinjectwould park the plugin in PENDING forever on deployments without those siblings. - The report contract has one owner.
AssembleReportRequest/AssembleReportResult/EvidenceInput/ReportSectionInputare imported as types fromdsh-research-report; they are never mirrored here, because a hand-kept mirror drifts silently.dsh-research-reportis a type-only devDependency (the build erases it —lib/index.jsmust never import it), and the untypedctx.getresult is proven by theisResearchReportLikeguard before use, never asserted withas unknown as. - Cordis events, not session-log events.
industry-research/*events are typed Cordis observability events and are never appended to the session log: the durable record is the workspace artifacts, model-visible tool results ride the durabletool/resultsession event, and observability rides the typed Cordis events. - No network outside
ctx.web. All retrieval goes through the official seam (provider selection, timeouts, and error taxonomy have one owner);offline: truedisables it entirely. - No invented data. A metric value without a source is a validation error; a missing value is an explicit gap slot; reports list gaps instead of filling them. Fictitious fixture data is clearly marked as teaching material.
- Workspace containment. Industry/company names are validated segments; data files are containment-checked against the session cwd (both sides
resolve()d — Windows backslash trap). - No tunables hardcoded. Every knob is a validated
Configfield with a default insrc/config.ts, an inline comment incordis.patch.yml, and a row in the five-language README configuration table. - Research-only compliance. Tool descriptions, cards, and reports carry 「仅供研究,不构成投资建议」; data points carry asOf and sources.
- This plugin registers no waterfall listeners. If one is ever added, allow/passthrough MUST call
next().
Checks
pnpm run typecheck && pnpm run typecheck:ci && pnpm run typecheck:checkout && pnpm test && pnpm run build && pnpm run verify:self-contained && pnpm run verify:artifacts && pnpm run verify:readme-sync && pnpm run verify:skills && pnpm run check:lockfile && pnpm run pack:check
typecheckresolves@deepseek-ai/*through the installed 0.1.7-rc.2 peers;typecheck:ciclearsskipLibCheckand enablesverbatimModuleSyntaxagainst the published types;typecheck:checkoutcompiles the same file set against the local harness checkout's built types throughtsconfig.checkout.json(no paths in the other two, so this is the only ruler that sees the checkout). All three must stay green.test/composition.spec.tsdrivesscripts/loader-runner.mjsin a child process.cordis-plugin-loader1.0.6 no longer makesloader.await()reject when a row'sapplythrew (Entry._reload()swallows the throw intofiber._errorin the same turn it resolvesfiber.inertia), so the runner re-awaits everyFAILEDrow —Fiber.await()still rethrows the stored error — and rethrows the first failure. Without that walk the invalid-config and default-export negatives are judged on the downstream symptom ("tool is missing from the tools registry") instead of the cause. The two reports are mutually exclusive — the rethrow precedes the registry check — and the suite asserts exactly that, so a symptom-only stderr fails.DSH_LOADER_RUNNER_NO_RETHROW=1disables the walk for re-measurement only.
Release
The release session bumps package.json + src/version.ts, stamps the CHANGELOG [Unreleased] section, re-runs the full gate, and commits + tags (never pushes from a dev session). .github/workflows/release.yml publishes to npm with provenance on version tags.
Docs
- Five-language READMEs (
README.md,README-zh.md,README-es.md,README-pt.md,README-hi.md) — keep all five in sync; the English file is the source of truth.scripts/check-readme-sync.mjs(CI) enforces section structure and configuration-table keys. - GitHub topics
dsh,dsh-plugin,deepseek-harness,cordis,industry-research,company-research,research,report(mirrorpackage.jsonkeywords). - License is Apache-2.0 (
LICENSE+ the package.jsonlicensefield).THIRD_PARTY_NOTICES.mddocuments the build-time dependencies.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

