oh-my-dsh-omo-fork-chs
The Simplified-Chinese unofficial fork of the tri-agent agent preset, packaged for
DeepSeek Harness as a publishable Cordis
bundle. Its preset id is omo-fork-chs and its order is 11, so it sits beside tri-agent
rather than replacing it: this package does not change the default preset and does not change
which preset a session picks. Only the persona prose is translated; the toolchain composition and
the orchestration doctrine are the upstream design.
The preset gives an agent a two-mode identity and a full coding toolchain:
- Atlas is the standing orchestrator. It does not implement, verify, or review; it reads the
plan, the notepad and the evidence, does the bookkeeping, and delegates every unit of work to a
named subagent.
- Prometheus is the planning doctrine. While a session is in plan mode (
/plan), the identity
swaps to Prometheus for the whole planning session and then swaps back. A single plugin,
plan-aware-persona.mjs, owns the one deployment:persona-prefix slot and chooses the text per
assembly by reading the public plan session projection. The planMode service is deliberately
not injected: it lives inside the isolate: { planMode: true } realm, so injecting it from
outside would never resolve.
Around that identity the preset mounts the coding toolchain: filesystem and shell tools, background
jobs, skills, goals, plan mode, compaction, delegation and workflows. The declaration carries 40
- id: rows, of which 9 are anchored subagent rows (explore, librarian, oracle, momus,
multimodal-looker, sisyphus-junior, metis, hephaestus, sisyphus). The package ships 11 persona
files; the plugin reads atlas.md and prometheus.md itself, and the other 9 are anchored to the
subagent rows.
What ships
| File | Purpose |
|---|
cordis.patch.yml | The published bundle patch: the preset declaration (id: preset-omo-fork-chs, config.id: omo-fork-chs, config.name: 'Oh My DSH Chs - OMO Fork', order: 11) and its plugins: list. 540 lines, 11 rewritten sites. |
plan-aware-persona.mjs | The package entry that registers the plan-aware persona prefix and the configured suffix. Exported as ./persona. |
index.js | The package root. It is intentionally empty: the bundle's behaviour lives in cordis.patch.yml and in ./persona. |
personas/ | 11 persona documents, translated into Simplified Chinese. |
baseline/ | A frozen snapshot of the pre-port patch plus a sidecar that explains what it does and does not prove. |
tools/ | The port, verification, baseline and package-check CLIs. |
test/ | The zero-dependency test suite, run with node --test. |
LICENSE-MIT.md | The MIT terms that cover our own work. |
LICENSE-SUL-1.0.md | The verbatim upstream Sustainable Use License 1.0 terms, which cover the upstream-derived content. |
NOTICE.md | The modification notice, the upstream attribution, and the machine-readable license scope mapping. |
The patch mounts the following peer packages, which DSH provides:
@deepseek-ai/dsh-agent-preset, dsh-agent-instructions, dsh-tool-bash, dsh-tool-pwsh,
dsh-tool-fs, dsh-tool-fs-search, dsh-tool-jobs, dsh-skill-filesystem, dsh-tool-skill,
dsh-command-goal, dsh-tool-goal, dsh-plan-mode, dsh-compaction-basic, dsh-command-compact,
dsh-compaction-tool-result-pruner, dsh-tool-subagent-control, dsh-tool-subagent,
dsh-workflow-ptc, dsh-tool-workflow, dsh-tool-ralph, dsh-tool-ask-user, dsh-tool-todo,
dsh-tool-web, dsh-tool-present, plus @deepseek-ai/cordis. That is 25 peer dependencies in
total.
Install
Installation is two explicit steps. Step 1 installs the dependency; step 2 tells the profile to
load it. Both are required, and dsh is not assumed to be on your PATH.
Step 1: install the package into the profile
Point the CLI at your DSH checkout's bin file, and forward the install to the profile:
node "<dsh bin>" plugin --profile <profile> add -w @tacrine_f/oh-my-dsh-omo-fork-chs
For example, on a checkout whose runtime is <dsh runtime>, <dsh bin> is
<dsh runtime>\@deepseek-ai\dsh\lib\bin.js, so:
node "<dsh runtime>\@deepseek-ai\dsh\lib\bin.js" plugin --profile web add -w @tacrine_f/oh-my-dsh-omo-fork-chs
This runs pnpm inside $DSH_HOME\profiles\<profile>\, so it writes a dependencies entry. It
does not touch dsh.profile.bundles.
Step 2: add the package to dsh.profile.bundles and restart
Open $DSH_HOME\profiles\<profile>\package.json and add the package name to the profile's
dsh.profile.bundles array:
{
"dsh": {
"profile": {
"bundles": [
// ...your existing bundles...
"@tacrine_f/oh-my-dsh-omo-fork-chs"
]
}
}
}
Then restart the Host and start a new session. A live agent keeps the composition it booted
with, so an already-running session will not pick the preset up; a new session is required.
Verify the install composed
$env:DSH_HOME = "<dsh home>"
node "<dsh bin>" --profile <profile> --dump-config
The output should contain a preset-omo-fork-chs row whose name is
@deepseek-ai/dsh-agent-preset, whose config.id is omo-fork-chs, whose config.name is
'Oh My DSH Chs - OMO Fork' and whose config.order is 11, plus nine personas expressions
resolving through this package. --dump-config prints the raw !!js expression text, so the nine
anchors appear as the resolve('<pkg>/package.json') form rather than as resolved filesystem
paths.
Post-install edits you may need
The published patch ships only the portable preset declaration. Two things that were local to
the machine this fork came from are deliberately not shipped, so you may want to re-add them
yourself:
-
Making omo-fork-chs your default preset. The published patch does not set a default
preset, and this package never forces one. If you want this preset to be the default, set the
registry row yourself in your own profile or overlay patch:
- id: agent-preset-registry
config:
default: omo-fork-chs
(@deepseek-ai/dsh-web-app inserts that row with default: standard; the row above replaces the
whole config, so only the default changes.)
-
Giving subagent_multimodal_looker a model route. The bundle this fork was derived from
carried a machine-specific provider and model route for that row, and that route is not
portable, so the published patch carries no agentOptions row at all. Without one the
multimodal subagent inherits the deployment's route. To pin one, add an agentOptions block to
that row in your own overlay:
- id: tool-subagent-multimodal-looker
name: '@deepseek-ai/dsh-tool-subagent'
config:
# ...the shipped config...
agentOptions:
provider: <your provider>
model: <your model>
Failure mode: it fails loud on purpose
plan-aware-persona.mjs reads its persona prose from disk at mount time. A missing or empty
persona file fails the mount, naming the file in the error:
plan-aware-persona: cannot read persona "file:///.../personas/atlas.md"
plan-aware-persona: persona "file:///.../personas/atlas.md" is empty
That is deliberate. An empty persona prefix would silently strip the agent's identity, so a mount
error that names the file is far easier to act on than an agent that quietly lost its instructions.
Peer model
Every @deepseek-ai/* package the patch mounts is declared as a peerDependency, never as a
regular dependency. DSH provides them; the bundle must not install its own copies. Peer ranges are
declared as >=0.1.7-rc.2 <0.2.0 || >=0.2.0-0 <0.3.0-0 || >=0.2.1-alpha.1 <0.3.0-0, which keeps the
0.1.7 line and adds the 0.2 line (and @deepseek-ai/cordis as ^4.0.1). The package declares 25
peers and no runtime dependencies and no dev dependencies, so nothing has to be installed
for the tests or for the zero-dependency part of the check chain.
The DSH profile's .npmrc sets auto-install-peers=false, so these peers are declarative only:
pnpm will not fetch them, and a missing peer surfaces as a resolution error at mount time rather
than being silently installed.
Verification
Everything below runs from the repository root with no install, except check:full, which needs
a runtime and is marked as such.
# 1. The test suite. Zero dependencies. 55 tests.
npm run test
# 2. The check chain. Zero dependencies, no runtime needed.
npm run check
# 3. The same check with the deep YAML comparison enabled. Needs a runtime.
npm run setup:verify
$env:DSH_RUNTIME = "<absolute path to a node_modules holding js-yaml and @deepseek-ai/cordis-plugin-include>"
npm run check:full
# 4. Regenerate the frozen baseline from the published patch and check it.
npm run baseline
npm run baseline -- --check
# 5. Re-port the published patch from a source patch directory.
npm run port -- --src <dir holding cordis.patch.yml>
What each one proves:
npm run test runs the node:test suite: the three forward rewrites, the inverse rewrites
and their round trip, the argument parsers, the machine-path pattern from both sides, the
zero-byte guard, and the package check's negative paths. It is 55 tests and it needs nothing
installed.npm run check is node tools/check-package.mjs && node tools/verify-patch.mjs. The first
runs 21 structural and license assertions (manifest identity, the four exports, the single
self-reference row, the nine persona anchors and their files, the files[] coverage, the eleven
persona byte counts, the license pointers, the three license documents, the paths that must never
ship, the untracked-file check over every directory files[] lists and the untracked root files
npm auto-includes, the upstream fingerprint
when one is supplied, the modification-notice claims, the license scope mapping and the per-entry
anchors
that pin each scope block to the content it claims, the machine-path
scan and the classes that govern where the pre-port package name may appear, the SHA256 of the
two modules the package copies out of this repository, the internal markdown anchors of every
document that carries one, and the patch's
shape). The second proves the published patch
still equals what the three documented rewrites compute from the frozen baseline, and reports
verdict: "IDENTICAL" with 11 rewrite sites. Both are zero-dependency, so they pass in a
clean clone with no install.npm run check:full is node tools/verify-patch.mjs on its own. With DSH_RUNTIME pointing
at an absolute node_modules that holds js-yaml and @deepseek-ai/cordis-plugin-include, the
verifier parses both documents with the real entry-list dialect and compares them structurally,
reporting yamlCompare: "ok" and sites.length of 11. Without a runtime it reports
yamlCompare: "skipped" and still exits 0, which is why the exit code alone is not evidence
here: assert on the field. That is also why this step is separate from check, and why
npm run setup:verify exists: it installs the runtime into .dsh-verify/ without adding a
dependency to this package. The tool reads DSH_RUNTIME itself rather than taking a --runtime
flag, because Windows npm runs scripts through cmd.exe and POSIX shells expand variables
differently, so one script string would behave differently on the two platforms.npm run baseline re-derives from the published patch by
applying the inverse of the three rewrites, and compares the committed baseline
against a freshly derived copy, printing the differing lines and exiting non-zero on drift.
npm run check and npm test must run inside a git checkout. Three of the package check's
assertions (the license scope
mapping, the machine-path scan, and the classes that govern where the pre-port package name may
appear) are defined over the set of tracked
files, so the tool asks git ls-files for that set. If git cannot answer, the tool fails with
run inside a git checkout rather than silently skipping those three. npm test needs the same
checkout for the same reason: its last case runs the package check against the real tree, so
unpacking the tarball outside a repository fails there too. A consumer who unpacks the
published tarball outside a repository will therefore see that error, and it is deliberate: a
mapping check that reports success for work it never did is worse than no check at all. The
published package includes the tools and the test suite so the chain can be re-run from a clone.
Counting a document's lines in PowerShell: Measure-Object -Line counts non-empty lines
only. (Get-Content README.md | Measure-Object -Line).Lines reports 311 for this file,
while (Get-Content README.md).Count reports 374 — the same file, two readings, and the
difference is 63 blank lines that the first command never saw. Nothing in the command's name or
output says so, so a length check built on it silently under-counts any file with blank lines in
it. Use .Count, or [IO.File]::ReadAllLines(path).Count, when you want the raw line count. The
same trap has already produced wrong readings of a plan file in this repository twice, from two
separate contributors, because the two numbers look equally plausible when you have not measured
both.
License
This repository is dual-licensed, and the two licenses are not alternatives you may choose
between: each file is covered by exactly one of them. It is not MIT-only.
The scope mapping, matching the two blocks in NOTICE.md:
LICENSE-SUL-1.0.md covers the upstream-derived content:
personas/, cordis.patch.yml, plan-aware-persona.mjs, index.js and
baseline/cordis.patch.yml.LICENSE-MIT.md covers our own work: tools/, test/, .github/,
baseline/README.md, package.json, .gitignore, .gitattributes, README.md, README.zh.md,
CHANGELOG.md
and NOTICE.md.NOTICE.md is the authoritative declaration: the modification notice, the upstream
attribution, and the machine-readable scope mapping, including the match rule that decides which
block covers each tracked path. In any disagreement between this README and NOTICE.md, the
notice wins.
The upstream terms are the Sustainable Use License 1.0, held verbatim in
LICENSE-SUL-1.0.md. SUL-1.0 is a source-available license. It is
not an OSI-approved open source license, and this repository does not claim otherwise.
The upstream terms permit use, copying, distribution and derivative works only for your own
internal business purposes, or for non-commercial or personal use, and they permit distribution
only if you do so free of charge and for non-commercial purposes. This package is
distributed under exactly those terms: free of charge, non-commercial, and not for resale. The
upstream license is not sublicensable, which is why the MIT license covering our own files
cannot be extended over the upstream-derived ones, and why LICENSE-SUL-1.0.md must travel with
every copy.
One carve-out applies to the MIT scope: the files under tools/ and test/ are our own work and
are licensed under MIT, but they embed verbatim fragments of the upstream bundle patch expressions,
quoted so that the patch can be generated and verified. Those quoted fragments remain subject to
the upstream SUL-1.0 terms and are not relicensed by the MIT license that covers the surrounding
code. The same carve-out is stated in NOTICE.md, and the two documents agree.
This repository is an unofficial fork. It is not affiliated with, endorsed by, or sponsored
by the upstream project or its author, and nothing here implies any affiliation or endorsement.
Upstream names and trademarks belong to their owners.
Known issues
- The declared peer range covers the 0.1.7 line and the 0.2 line, and was verified on one 0.2.x build. The 24
@deepseek-ai/dsh-* peers are declared as
>=0.1.7-rc.2 <0.2.0 || >=0.2.0-0 <0.3.0-0 || >=0.2.1-alpha.1 <0.3.0-0, which keeps the range this
fork inherited from upstream and adds the 0.2 line. The host decides compatibility with a semver
range check over those peers, evaluating prereleases, so the range accepts 0.2.0,
0.2.1-alpha.1 and the later prerelease 0.2.2-beta, and rejects 0.3.0-alpha.1.
Verified by experiment on 0.2.1-alpha.1, which is the only 0.2.x build available to us: with
no exemption present in the profile, a config dump of this package exits 0, writes 0 bytes to
stderr, and carries the preset with all nine persona anchors. That is a claim about the mount and
the anchor paths, not about the personas' bytes and not about the preset's behaviour: no agent was
invoked, and a persona that still existed but had been truncated would pass the same check.
0.3.0-alpha.1 is the re-entry condition: a future runtime in that range is rejected again,
with the same incompatibility warning on stderr and the same exit 0, so the failure stays silent
unless you read stderr. On such a version you must either run a dsh inside the declared range, or
grant the exact-version exemption with
dsh plugin allow-version "<pkg>@<version>" --dsh-version <version> --accept-risk. That exemption
is the host's own escape hatch and carries the risk the host itself warns about, so it is a
deliberate acceptance of risk rather than a supported configuration. It is still needed for cases
this widening does not cover: the published 0.1.0 on npm carries the old narrower range, and any
future 0.3.x runtime will need it. @deepseek-ai/cordis is declared at ^4.0.1 as well, but the
host never tests it — it only evaluates peers named @deepseek-ai/dsh or @deepseek-ai/dsh-* —
so that entry is not a compatibility constraint the host enforces.
baseline/ proves reproducibility, not correctness. The baseline is a frozen snapshot of the
pre-port patch, committed so the round trip can be checked anywhere. It is byte-identical to the
source patch, and the derivation proves the port is reproducible: the published patch and the
baseline are related by exactly the three documented rewrites and nothing else. It does not
prove the content is correct. If the source patch had been wrong, the derivation would faithfully
reproduce that wrongness and every check here would still pass.- The prior evidence for content correctness is not shipped. That evidence is the
evidence set produced by the Simplified-Chinese port work, whose
checker reported and whose
checker reported . Those checkers are : they run against directories
that exist only on the machine the port was prepared on, and they are deliberately not shipped
with this package.
Lineage
This package descends from the Simplified-Chinese agent-preset bundle, which lived as a link
install: a private dependency wired into one machine's DSH profile, pointing at a local directory.
That form cannot be installed anywhere else, because its patch named a machine-local package and its
persona anchors resolved relative to the profile directory.
This repository turns that bundle into an npm-installable form. The three documented rewrites do
the work: the package self-reference and the nine persona path anchors are re-pointed at this
package's own name and its package.json, and the header comment follows. The result is a patch
that resolves in any install layout, a frozen baseline that makes the round trip checkable without
the original machine, and a check chain that proves the published patch still matches the source it
was derived from.
The upstream project is oh-my-openagent, formerly published as oh-my-opencode
(https://github.com/code-yeongyu/oh-my-openagent). The content in this repository is derived from
upstream commit 3c711483856207ef530c09d062fa9077c225debbe, which carries upstream version 5.0.1.
The persona prose, the toolchain composition and the plugin are the upstream design; this fork's
contribution is the translation into Simplified Chinese and the packaging split.
