dsh-raven-research
DeepSeek Harness plugin for progressive, source-grounded deep research, writing, and learning: steerable checkpoints and citations verified against retrieved sources
71 results
DeepSeek Harness plugin for progressive, source-grounded deep research, writing, and learning: steerable checkpoints and citations verified against retrieved sources
AI Agent runtime authorization & evidence verification — tool-call GuardrailProvider, CCS 7-dimension verification standard, MCP/DSH security scanner, SSRF/command-injection/credential-exfil blocking with Ed25519 signed receipts.
Talk to Excel in DeepSeek Harness: create/edit spreadsheets, formulas, styles, filters, and tables by conversation; auto-verifies formulas after every edit, repairs silent errors, and validates charts.
Read-only browser verification tools for the DeepSeek Harness web GUI: browser_open / browser_mock / browser_assert / browser_screenshot — verify a page (H5/desktop) in ≤4 tool calls with mock interception, DOM assertions, and screenshots that auto-project into the model context.
Portable Agent Notes mechanism: verification gates, scaffolding CLI, maintenance skills, and the AN preset for dsh
Computer Use for DeepSeek Harness backed by the cua-driver daemon (trycua): accessibility element-level targeting, background-first input delivery, window/desktop screenshots, deterministic verification.
Evidence-first inspection, compatibility verification, and quarantine tooling for DeepSeek Harness plugins.
Performance diagnosis, repeated plugin isolation campaigns, safe recovery, and measured verification for DeepSeek Harness
Evidence-gated frontier mathematics research workflow for DeepSeek Harness
Completion verification for DeepSeek Harness: gates goal stamps (update_goal complete) and turn boundaries — a goal stays open until workspace verification commands exit 0
A blind, fair, local Agent arena inside DSH Web: same task, same commit, isolated worktrees, shared verification, judge before you reveal.
Native DeepSeek Harness plugin: a turn that states a verdict without evidence does not end — it is steered back for proof.
Infrastructure plugin for dsh: provisions the kimi-webbridge daemon (auto-install with SHA-256 verification, auto-start) and registers the kimi-webbridge skill that documents the full 25-action browser protocol. Registers a thin kimi_webbridge tool (status health check only).
Natural project teaching and first-delivery verification guidance using native DSH instructions.
DSH skill bundle: competition math (IMO/Putnam/USAMO/AIME) solved with a pure-reasoning pass, an adversarial verifier in a fresh subagent context attacking concrete failure modes, and calibrated confidence output (high / medium / honest "no confident solution"); optional LaTeX→PDF rendering.
Monorepo-aware execution planning, Harness Jobs handoff, and verification gates for DeepSeek Harness
Cloudflare Access JWT verification and remote DSH privileged authorization
Mechanically stops unfounded "done" claims from coding agents — a zero-LLM, pure-regex turn-boundary hook that demands verification evidence before sign-off.
Zero-dependency static + sandbox smoke detector for DeepSeek Harness (dsh) plugins: package-structure gates (R), cordis contract scans (K), keyless-headless sandbox smoke (D), and ecosystem-listing checks (CC).
Verifiable authorization, execution evidence, and offline delivery verification for DeepSeek Harness
DeepSeek Harness plugin: PDF→Word (.docx) conversion with layout fidelity (fonts/tables/images/borders), OCR scan mode, and optional multimodal LLM verification. Registers the pdf_to_word model tool.
Set up free, encrypted, deduplicated backups of remote Linux servers, pulled from a Windows PC with Restic — tar-over-SSH streaming, zero software installed server-side, with size verification against
Strict check: verify code and commands against real checkers instead of reading them — Lean 4 kernel checking with an axiom audit, language type/syntax checks, and static defect rules.