Watches DSH and plugin releases, retests exact published artifacts in disposable runners, publishes machine-readable compatibility evidence, and reconciles managed issues after fixes.
Install
# from npm (prebuilt)
dsh plugin --profile web add upstream-radar
# from GitHub (first run asks for allowBuilds approval — follow the hint, retry)
dsh plugin --profile web add github:MicroMilo/upstream-radar
Any plugin you install runs third-party code with your own permissions — it can read your files, use your credentials, and reach the network, and tool approvals don’t sandbox it. GitHub-sourced plugins also run build scripts at install time — pnpm blocks those until you allow them, so an install can stop with ERR_PNPM_GIT_DEP_PREPARE_NOT_ALLOWED or ERR_PNPM_IGNORED_BUILDS; dsh prints the exact key to add under allowBuilds in your profile’s pnpm-workspace.yaml, and the install works on the next run. Allowing a build is a trust decision: only install sources you trust, and pin a commit (github:owner/repo#sha).
README
Upstream Radar watches exact DSH and plugin releases, persists every affected case, and wakes an Agent only when the input changes. The Agent reads the repository, chooses evidence-backed Node.js and execution profiles, installs and runs the exact artifacts in disposable CI, watches them while they run, and publishes a reviewable compatibility report.
No guessed matrix. No “install succeeded, therefore compatible.” No silent unknowns.
What you get
- Change-driven analysis — a DSH release fans out to your plugin cohort; a plugin release analyzes only that plugin.
- Agent-owned environment discovery — README, manifests, lockfiles, CI, and startup scripts determine Node.js, package manager, and profile selection.
- Active runtime supervision — the Agent reads incremental logs, inspects running processes, handles recoverable build gates, adjusts, and retries.
- Exact, durable evidence — every report is bound to plugin bytes, source commit, DSH version, Node.js version, profile, commands, and logs.
- Safe retries and deduplication — interrupted executor work stays pending; an unchanged completed input does not run twice.
The loop
flowchart LR
Change["Schedule or upstream change"] --> Task["Persist exact task"]
Task --> Agent["Agent reviews repository evidence"]
Agent --> Run["Install, run, watch, recover"]
Run --> Report["Versioned report and logs"]
Report --> State["Deduplicate or retry"]
State --> Change
The model recommends and operates within a bounded tool contract. Deterministic code selects exact versions, verifies artifacts, isolates execution, and decides whether evidence is complete. Plugin code never receives model credentials or repository write tokens.
See it working
- Live compatibility report index — completed means the analysis closed, not that the plugin passed.
- A real
contextAgent run — repository review, Node/profile reasoning, isolated execution, and active watches in one case. - A green durable retry cycle — the same persisted inputs produced an empty matrix instead of duplicate work.
Reports preserve compatible, incompatible, unknown, and externally blocked
outcomes separately. Missing credentials, missing coverage, and executor faults
cannot become a pass.
Try it in 30 seconds
Inspect one exact published artifact without executing plugin code:
npx --yes upstream-radar@0.45.0 inspect \
@sanqi-normal/dsh-webui-market-plugin@0.5.4 \
--deep --fail-on never
Or review a public plugin repository without installing it:
npx --yes upstream-radar@0.45.0 scan \
https://github.com/owner/dsh-plugin \
--fail-on never
The static commands read bounded package and repository evidence. They do not install dependencies, execute lifecycle scripts, start DSH, or call an LLM.
Automate the compatibility loop
Start from one maintained workflow:
- Watch a plugin repository every day
- Review an exact plugin across DSH versions
- Gate dependency changes in CI
This repository's deployed loop runs from
upstream-observer.yml. Its single
operator entry point is
examples/dsh/active-agent/policy.json:
{
"schema": "upstream-radar.dsh-active-agent-policy/v1alpha1",
"dsh": { "channel": "next" },
"defaults": {},
"plugins": [
{ "targetId": "context" },
{ "targetId": "dsh-tui" }
]
}
Leave Node.js and profiles unset to let the Agent infer them. Override them
globally or per plugin when you need a fixed experiment. Exact DSH or plugin
versions require a matching sourceRef, so repository evidence cannot drift
from installed bytes. See the policy reference.
What a result means
| Result | Meaning |
|---|---|
compatible |
The exact tested cell completed its required install and runtime checks. |
incompatible |
Reproducible evidence failed a required compatibility boundary. |
unknown |
Execution happened, but coverage or attribution was insufficient. |
blocked |
An external account, credential, source, or executor prevented completion. |
Results are scoped observations, not permanent compatibility badges or security certificates. A new artifact, source commit, DSH release, runtime policy, or expired evidence creates a new input.
Built for
- Plugin authors who want release failures caught before users report them.
- DSH ecosystem maintainers who need one comparable, auditable evidence feed.
- Platform teams that need repeatable upgrade decisions instead of a manual README-and-log investigation.
Upstream Radar is listed by awesome-dsh-plugin, awesome-deepseek-harness, and awesome-deepseek-harness-plugins.
Links
More in this category
toby-bridges/api-relay-audit★ 861
Runs local security audits of AI API relays and LLM proxies from DeepSeek Harness, producing Markdown reports for prompt injection, model substitution signals, tool-call rewriting, error leakage, stream integrity, and profile-gated Web3 risks.
SeaOf0/dsh-redteam-model★ 646
Authorized-security DSH collection: nine work modes (redteam coordinator, pentest, code audit, binary analysis, attack-defense, AV evasion, incident response, cloud security, CTF solving) and fifteen runtime plugins, managed from a settings page with one-click deploy, install, update and uninstall.
howmp/dsh-pentest★ 559
Authorized pentest mode for DeepSeek Harness — exploration chain, assets and findings with a Web view.
PerryLink/dsh-auto-review★ 212
Second-model auto-review on the approval answerer chain: a read-only reviewer subagent returns structured allow/deny verdicts with reasons, fail-closed by default.
NanmiCoder/dsh-auto-mode★ 164
Adds an Auto permission preset between Workspace Write and Full access: routine work stays in the official workspace-write sandbox while the current session model reviews escalation and destructive calls, granting one exact wider access once, asking when the intent is ambiguous, and denying critical paths.
PerryLink/dsh-permission-rules★ 115
Claude Code-style declarative permission rules: ordered allow/deny/ask YAML rules matching tool names, arguments, workspace paths, and agent identity on the tools/pre-execute waterfall, with full session-log audit, dry-run mode, and hot reload.
Community comments
Comments are public GitHub Discussions. Loading them connects to GitHub and Giscus; a GitHub account is required to post.