Plugins

One repository, three parts: a runtime injector for DSH plugin packages (inject, hot-reload, unload, promote a dev staging tool to the front, route self-heal, plus a settings-page plugin manager that lists, unloads and drags folders in to internalize), a task-aware reasoning-mode router agent preset (router-standard / router-spec / router-react), and a graded two-level task protocol whose six tools (commit_star, lock_stage, revise_do, edit_plan, mark_task, redteam_verdict) pin task state to disk. The injector implementation ships in-tree, so the install carries its own behaviour rather than a dependency list.

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.

Verifiable research-report engine for DeepSeek Harness with a content-addressed evidence ledger, versioned sealed reports where every claim carries a verification verdict and the manifest hash seals the directory, and retrieval orchestration that reuses the ctx.web and ctx.jobs seams.

Red-team engagement mode: send one target organization name and a planner session runs a five-role execution team (recon, asset triage, vulnerability discovery, exploitation, internal pivot) at up to three concurrent agents, preflight-checking skills and resources and asking once for any missing key or VPS first, guided by a built-in first-run onboarding skill; scoring follows the merged intrusion scoring rules (8 categories / 25 points) with server-side caps, highest-privilege-wins and per-service dedup, and self-registered accounts never score; findings land in a local SQLite fact base with discovery timestamps, shown in a persistent right-side console of 12 tabs (asset mapping with a discovery timeline, agent roster, session/tunnel state, five-stage attack chain, scoring targets, a report that spells out how each score was obtained — actions, exact commands, credential provenance, tunnel build commands — a category-organised POC/EXP knowledge base and a skill library with per-skill usability verdicts); ships 23 native skills, 53 redteam_* tools and one-command self-update.

White-box causal guardrail for dsh: every tool call is adjudicated before execution along a deterministic causal logic chain — blocks destructive and credential-file operations, escalates repeated boundary violations, cuts faulting links until verified-fixed, and gives risk verdicts on un-auditable execution; runs fully local with zero API cost, plus 6 white-box self-check tools.

The complete #dsh-plugin mall inside DeepSeek Harness: browse, search and score the full GitHub dsh-plugin topic catalog refreshed every 2 hours, AI pre-install review with install/caution/avoid verdicts, five-dimension scores with radar charts, nine-language UI, editor picks and personalized recommendations, one-click install/update/rollback with a global task panel, and an in-agent /dsh-mall skill.

Spec-Driven Development pipeline for Odoo: 15 tools take a request from an immutable spec through architecture to module code or live configuration, then require a persisted verdict backed by evidence from a real Odoo instance, with fail-closed gates, human-approved batches in a declared environment, Odoo's own CSV/Excel importer, and a journal that reverses the data written through the plugin.

Plugin trust pipeline for DeepSeek Harness: deterministic static scan with verdicts, opt-in runtime guard with honeypot lures, agent audit-protocol skill, and a browser shield status light. Alarm-only, never an enforcer.

English edition of Weiwen's Law (KISS's Law): the first and only domain-agnostic white-box causal adjudication middleware for AI agents. Unlike domain-specific causal-inference tooling (statistics/economics/ML) or evidence-gated agent firewalls, it adjudicates *every* tool call along a deterministic structural causal chain (R->S->D->H->M) before execution - across any domain, with no training or per-domain tuning. It blocks destructive and credential-file operations, escalates repeated boundary violations, cuts faulting links until verified-fixed, and gives risk verdicts on un-auditable execution; runs fully local with zero API cost.

Runs a thesis-antithesis-seal debate whose verdict is computed from real command exit codes and case counts, never from agent prose.

Static capability disclosure for DeepSeek Harness plugins: a Settings page and CLI that list installed plugins' capabilities, literal destinations, install scripts, and injection shapes (prompt registrations, shipped skill text, bundle patch), each with file:line evidence. Code-determined, reproducible, zero-token. Disclosure only — not a security verdict, never a safety claim.

Independent browser acceptance testing for agent deliverables: JSON spec in, real-Chromium verdict out.

21 read-only lifecycle governance tools in one self-contained DSH plugin: task tiering, state/plan/evidence validation, gate precheck and aggregation, release/install audits, context and progress analysis, critical path, resume reconciliation, and a mutation preview that always denies execution. Installs from npm as @240xu/dsh-tech-lead.

Security-audit facts AI coding agents cannot compute, plus a verdict gate: euthyna history attributes deleted lines to commits and flags those from security fixes (--origins finds the first introducer), euthyna coverage reports which changed symbols no test ever invoked, and euthyna gate checks each finding against six gates, downgrading any without evidence to an observation.

Adds an Explain button to the approval card: one model call returns exactly three lines reading the pending operation, covering what it does, which files or system state it reads, writes, modifies or executes, and a risky/attention/safe verdict with a one-sentence justification. It also renders the detail area for `write` and `edit` calls, which the shipped renderer leaves blank because it only reads a `command` field.

Content-degradation detection (thinking loops, output repetition) for DeepSeek Harness, with a continue-vs-handoff cost verdict and zero-model handoff documents.

Skill that fact-checks a claim against the open web rather than model memory: it splits the claim into independently checkable fact points, routes each point to the primary-source ladder for its domain (policy, academic, medical, financial, disaster and accidents, food safety, science, education, sports, entertainment, judicial, and more), searches domestic and international ecosystems in parallel, fetches primary pages, grades sources by type, re-searches with debunking keywords, and reports conclusion-first with a full URL and citation position behind every statement, keeping unverified and undecidable points separate from confirmed ones. Three factual judgements can be routed to a TypeSafe Jev (System One) decision model when a key is configured, falling back to the skill's deterministic rules when it is not; a model verdict can only lower a confidence tier.

Per-turn file changes, per-file diffs and deterministic safety verdicts for DSH sessions, plus a preview-first rewind that refuses when the worktree has drifted since the preview.

Mandatory cross-review for design proposals and lessons-learned: auto-detects design doc writes and enqueues them (tier=review) into the task queue, delivers the external reviewer verdict back to the initiating session, and appends approved preventive measures to OPS-GUARDRAILS.md (append-only, conflict-checked).

Answers hard questions by staging a formal debate: two adversarial subagents argue opposite sides with research, then the main agent rules on the transcript through the debate_open, debate_round, and debate_verdict tools.

Registers /wrapped, which reads the session's own stats and renders a shareable dark SVG summary card — tool vs. thinking time, decode speed, and a tool/thinking-ratio verdict.

Plugin insight center: one answer to "哪些值得装" — plugin_guide matches needs to plugins, recipe installs whole environments, plugin_rank scores health (0-100), plugin_audit static-scans a local checkout for security, and plugin_verdict outputs install / caution / research / avoid.

Verified restart for DeepSeek Harness: restart without privileges, check that the harness really came back, report the verdict to the session that asked, and hand back what was left to do.

Measure whether a dsh setup change actually helped: register repeatable cases, run them, and diff the results before and after you change rules, skills, or models.