agentleFS
wowyuarmwowyuarm/dsh-agent-team/AGENTS.md

dsh-agent-team

AGENTS.md89 starsChanged 9 days agoMITView on GitHub

dsh-agent-team is an opt-in, externally installable Cordis bundle for DeepSeek Harness. It adds a single-host Agent Team without modifying the Harness repository, the agent loop, or shipped defaults. Read docs/README.md first — it is the maintained documentation index. Use this authority order: Keep production code self-explanatory through clear names, types, and structure. Comments explain only non-obvious constraints, ownership, or reasons; they are not the sole definition of current behavior. The check:comments row in docs/development/start-and-checks.md holds the per-scope budget and the…

1 other project uses a version of this.

Adopted by

Most recently taken up by

  • Other versions1 project uses a version of this.

    Word for word · 1

    1. Contexera/dsh-agent-team★ 89100% same

    Edited · 0

    None found.

    Wrote their own version · 0

    None found.

    How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.

  • 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 31,843 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 repository12 other files it gives its agents.

The file

Show its rules and text

We split this AGENTS.md into the rules it gives an agent

16 rules. About 1 is common, 5 are distinctive, and 10 are very specific to this project.

  • ≈1

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈5

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 10

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  • could be automatic: a setting could enforce it
  1. Very specific:

    Read docs/README.md first — it is the maintained documentation index.

    line 7
  2. Distinctive:

    Use this authority order:

    line 9
  3. Distinctive:

    Keep production code self-explanatory through clear names, types, and structure.

    line 16
  4. Very specific:

    Use the docs/architecture/ pages for package ownership, Host authority, Remote, preset, and Client boundaries — not before every edit, only …

    line 22
  5. Very specific:

    Use docs/harness-navigation.md for cross-repository routes when a change consumes a Harness capability.

    line 23
  6. Very specific:

    Treat .scratch/ as background context, not as a specification to implement blindly.

    line 24
  7. Distinctive:

    Keep durable UI mutations non-optimistic.

    line 34
  8. Distinctive:

    Keep Team tools/guidance inside the isolated Team preset.

    line 35
  9. Very specific:

    Use ESM, strict TypeScript, .ts/.tsx local imports, existing package names, and existing branded domain types.

    line 39
  10. Distinctive:

    Keep one authoritative formatter/projection per domain concept.

    line 40
  11. Very specific:

    Do not hand-edit generated tsconfig*.json path facades.

    line 41
  12. Very specific:

    Update the affected maintained document, package README, and tests when a non-trivial change alters a public contract or workflow.

    line 42
  13. Very specific:

    Follow docs/AGENTS.md when editing or adding maintained documents — routing branches and the maintenance discipline live there.

    line 43
  14. Very specific:

    Write each commit message as one Conventional Commits subject line: type: lowercase imperative summary (feat, fix, chore, refactor, test, …

    line 45
  15. Common:

    Do not commit credentials, temporary profiles, browser overlays, generated test files, browser artifacts, or build residue.

    line 46
  16. Very specific:

    Run the narrowest applicable check, then escalate for the changed surface.

    line 50

≈ “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.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
AGENTS.md

Repository Guidelines

Scope and authority

dsh-agent-team is an opt-in, externally installable Cordis bundle for DeepSeek Harness. It adds a single-host Agent Team without modifying the Harness repository, the agent loop, or shipped defaults.

Read docs/README.md first — it is the maintained documentation index.

Use this authority order:

  1. Current behavior: source and tests under packages/. When prose disagrees with code, code wins.
  2. Maintained engineering guidance: docs/. Update it when a code or workflow change invalidates it; it must not define behavior that the code does not implement.
  3. Harness contract: ../deepseek-harness/docs/, its applicable AGENTS.md, and the corresponding Harness source/tests.
  4. Work history: .scratch/ — active work items and archived design, research, tickets, prototypes, and validation evidence. It is not an implementation or API authority; its own AGENTS.md governs how to work there, and the README.md indexes it. Verify relevant history against code, tests, and the Harness contract.

Keep production code self-explanatory through clear names, types, and structure. Comments explain only non-obvious constraints, ownership, or reasons; they are not the sole definition of current behavior. The check:comments row in docs/development/start-and-checks.md holds the per-scope budget and the rules a comment's references must satisfy.

When deleting comments in bulk, a green test suite proves nothing about the change. Tests assert behavior; comments are text. Measure the edit itself instead: audit the diff so every removed line is a comment or a blank line, and treat any removed code line as a failure. Delete whole comment blocks rather than merged line ranges — a merged range swallows the code between two comments — and never normalize whitespace outside the lines you removed. This applies to any sweep over comments, including one driven by a subagent report: reports carry line numbers that drift, so verify each cited range against the current file before acting on it.

Reading guidance

  • Use the docs/architecture/ pages for package ownership, Host authority, Remote, preset, and Client boundaries — not before every edit, only when a change crosses those boundaries: package-ownership.md, host-authority.md, tools-and-preset.md, client-and-remote.md, workspace-session-storage.md.
  • Use docs/harness-navigation.md for cross-repository routes when a change consumes a Harness capability.
  • Treat .scratch/ as background context, not as a specification to implement blindly. New cross-session work belongs in one .scratch/active/<work>/ directory; close it into the archive only after durable conclusions have moved to maintained documents.

Architecture guardrails

  • packages/agent-team is the only Team authority. Its append-only operation ledger is the durable source of Team facts; projections, tools, commands, Remote, and UI do not maintain parallel authority.
  • packages/tool-agent-team provides the nine model-facing Team tools only through the explicit team-member preset. The Web Client is the only Human Team control surface; do not restore a slash-command adapter.
  • Team is an external plugin. Do not modify ../deepseek-harness, its agent loop, or shipped defaults for ordinary Team work. If a public Harness extension point is insufficient, record the limitation and decide whether to implement a Team-owned plugin or change the design.
  • Typed Remote declarations are the input; Typert artifacts are generated. Never hand-edit packages/agent-team/lib/typert.*.
  • The Client uses public Harness plugin and slot APIs. Mount the generated Remote, wait for remote.agentTeam, and use ctx.slots.inject() when a declaration may not exist yet; dsh.client.inject is not an activation-order guarantee.
  • A slot parent’s children declaration is its render authority. Do not copy private shipped UI or redeclare sidebar.workspaces.directoryFlow from the Team workspace shadow; live duplicate child declarations are rejected by Harness SlotCore, regardless of priority.
  • Keep durable UI mutations non-optimistic. Authorization, membership, revisions, idempotency, and stable branded refs are enforced by the Host.
  • Keep Team tools/guidance inside the isolated Team preset. Ordinary Sessions must not acquire Team behavior accidentally.

Code and documentation conventions

  • Use ESM, strict TypeScript, .ts/.tsx local imports, existing package names, and existing branded domain types.
  • Keep one authoritative formatter/projection per domain concept. Do not add speculative registries, compatibility layers, fallback authorities, or duplicate stores.
  • Do not hand-edit generated tsconfig*.json path facades. Change scripts/sync-paths.mjs and regenerate.
  • Update the affected maintained document, package README, and tests when a non-trivial change alters a public contract or workflow. Do not update archived .scratch/ material merely to make it agree with new code.
  • Follow docs/AGENTS.md when editing or adding maintained documents — routing branches and the maintenance discipline live there.
  • For visible UI changes, run the applicable component checks and npm run test:browser, inspect the new artifacts/browser/ screenshots at desktop and 390×844, and verify keyboard/focus, dialog/menu accessibility, loading/error/empty states, and ordinary DSH restoration. Keep only a small, human-approved milestone set under .scratch/archive/.../validation/; routine screenshots stay ignored.
  • Write each commit message as one Conventional Commits subject line: type: lowercase imperative summary (feat, fix, chore, refactor, test, perf, docs), with no body. Match the existing git log style.
  • Do not commit credentials, temporary profiles, browser overlays, generated test files, browser artifacts, or build residue.

Checks

Run the narrowest applicable check, then escalate for the changed surface. The exact workflow is in docs/development/start-and-checks.md.

npm run test:browser is required for changes that can affect the assembled Web bundle, Client loading, slot takeover, Remote activation, or visible UI. It uses the adjacent Harness checkout and a temporary profile; it does not modify the Harness repository permanently.

Release cadence is batched: between releases the operator daily-drives a locally linked build as a lightweight acceptance channel, so choose the narrowest applicable check per change instead of demanding full acceptance for every small fix. When asking the operator to preview or accept a change, point them at the dev profile (dsh web --profile web-dev); the default web profile stays on the published stable release. See "Profile 模式与发布节奏" in docs/development/environments-and-install.md.

Further reading

  • Domain and Host behavior: packages/agent-team/src/, its tests, and docs/architecture/README.md.
  • Tools and preset: packages/tool-agent-team/, the team-member row in cordis.patch.yml, and the matching Harness subsystem docs.
  • Client or UI: packages/client-agent-team/src/client/, docs/architecture/README.md, and docs/harness-navigation.md; consult the relevant .scratch/archive/ work history only when needed, and verify all behavior in code and tests.

Download now (ZIP)

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
What did you use it for? e.g. Reviewing pull requests in a Rails app

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.