PerryLink

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.

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.

Industry and company research domain pack: industry_map chain maps, public-source policy/news tracking over ctx.web (industry_track), company_scan cards from user data files, and industry_report with an optional ctx.researchReport sealing bridge and a builtin-fallback renderer, plus two methodology skills.

Bounded, layered, approval-gated, auditable cross-session memory: a typed `ctx.memory` seam with a zero-dependency SQLite provider, a `memory` tool, and frozen snapshot injection, plus the dsh-memory-protocol v1 rehearsal with an adapter registry and a distributable conformance suite.

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.

Read-only runtime management panel for the official DSH MCP client: connection status, registered tools, errors, and reconnect counts through the /mcp command and a Settings tab, with sanitized display and enable/disable patch suggestions.

The DSH plugin-development knowledge base as an on-demand agent skill: official constraints, task workflows, API reference and community gotchas, installed with the bundle so the agent can look things up while building a plugin.

Deterministic research reports for Chinese public mutual funds built from public-source data (Tiantian Fund and Eastmoney), pure-function metrics (performance decomposition, holdings penetration, style attribution, manager profile), and versioned reports with a per-number snapshot traceability appendix.

Engineering-discipline guard: requirements grill before the first edit, red/green test-evidence gates, forked adversary review, and a delivery report with a per-dimension verification workflow.

Deterministic data profiling, cleaning, and verification: data_profile / data_clean / data_verify tools plus a frozen cross-plugin verifyCitations citation-checking contract, with durable reports in a storage domain.

Four-source migration wizard: move Claude Code, Codex, OpenCode and Hermes sessions, memories, skills, instructions and slash commands into DSH (/move wizard with approval gate and idempotent move.json, resumable sessions).

LSP action surface for DSH: diagnostics, formatting, completion, code actions, symbols, signature help, inlay hints and rename, all backed by real language servers.

Claude Code /rewind for DeepSeek Harness: git-first workspace snapshots before every mutating tool execution, turn-boundary session forks, and a one-shot /rewind command that restores files and forks the session back to a checkpoint.

Terminal-style input history for the web composer: edge-first arrow-key recall with exact draft/caret restore, browser-local persisted history, Ctrl+R reverse search, and sliding-context awareness; 0.5.0 adds a smart input layer — cross-session snippets (/save, /load), prompt templates with variables, reuse insights, and compaction-summary highlighting.

Durable background child agents on the official subagent seam: start from any session, watch progress in the Web UI sidebar, message and interrupt any time, with per-child tool scoping, persona and delegation-depth caps.

Ollama local-model adapter: ollama_list/pull/remove/show plus health checks, registers an Ollama route via the official LlmAdapter with model_route rules (offline-first / long-text / privacy) and cloud fallback; /ollama command for a one-shot overview.

Security-audit methodology skill pack plus the plugin_vet supply-chain gate: eight agent skills (secret scan, dependency audit, supply-chain review, prompt-injection review, audit orchestration, threat modeling, vuln intel, incident response) in Chinese and English editions, with an npm provider bundle that mounts the skills and registers the automated plugin_vet pre-install scanner.

Voice-first session loop: a composer microphone button with browser/local speech-to-text (Web Speech, FunASR, whisper.cpp), a speak tool for text-to-speech replies (browser, edge-tts, piper), event announcements with mute, and speak-to-interrupt.

Official-grade GitHub CI integration: a composite action.yml, a polling PR review bot with idempotent inline comments and a status-check gate, plus PR/issues tools with every write gated by human approval.

Pin sessions and workspaces to the top of the Web sidebar with per-pin row colors, a header toggle and a pinned panel; 0.4.0 adds a navigation organizer — pin groups (boards), tags and saved filter views, session health summaries, and /goto.

Turns local markdown and text documents into a queryable knowledge base with hybrid semantic and keyword search, citation verification, and source injection.

Exports the session event stream to OpenTelemetry OTLP and Langfuse as sanitized, buffered traces and metrics, off by default.

Runtime-switchable model output styles with Claude Code outputStyles parity, plus the output.render.* presentation protocol: a /style command, per-session persistence, systemPrompt injection, six built-in styles, a web picker, and a renderer registry with per-session/per-tool rules and /export.

Runs isolated install-and-smoke test drives for DSH plugins in throwaway profiles, returning structured pass/fail records and batch matrices without touching your real profile.

Detects prompt-injection, jailbreak, and secret-leak patterns on the agent/pre-step, tools/pre-execute, and tools/post-execute seams with allow/ask/block tiers, sanitized defend/detection audit events, a defend_report tool, and a destructive-delete command guard.

Cross-device session sync: a dedicated git mirror with append-only keep-both conflict resolution, a /sync command and sync_status/sync_pull/sync_push tools.

Cost governance for DeepSeek Harness: aggregated token/cost metering per model, session and day, session/daily/monthly budget caps with threshold alerts and alert/block/degrade over-limit policies, carbon footprint estimation, per-model latency benchmarks, a Settings budget tab, and the /budget command.

Unified static-image generation router: one image_generate tool with standard parameters, config-driven OpenAI-compatible engine routing (OpenAI Images, Zhipu CogView, and any compatible endpoint) with health-aware fallback, durable workspace attachment results, per-session quota accounting, an in-conversation result card, and a Plugins settings panel that stores API keys as credential references.

PII masking middleware: anonymizes names, phones, emails, ID cards, bank cards, keys, and addresses before the model boundary and restores placeholders at display; plaintext never enters the session log; /mask command plus mask_test tool.

Vendor parameter translation and deterministic JSON repair for DeepSeek Harness: a /translate command maps 13 canonical parameters across 11 vendors, and the post-execute repair layer (plus fix_json) fixes broken JSON tool output without fabricating data.

Read-only performance diagnostics: session load timing, spill hits, compaction stats, context-injection volume, and cache-hit rate via the /fast command and fast_report tool, sampled asynchronously off the model path.

Native desktop control for DeepSeek Harness (Windows first): screen_shot, screen_read accessibility trees, click/type/scroll/key, and app launch — approval-gated, never stealing foreground focus.

Multi-dimensional quality scoring for DeepSeek Harness plugins that scores a repo or npm package across install success, maintenance activity, documentation completeness, security scan, and protocol compliance using real CLI evidence, and produces a JSON or Markdown leaderboard report.

TickTick (Dida365) task bridge: a session-header task panel and curated agent tools over the official TickTick MCP endpoint.

Version-locked DeepSeek Harness plugin upgrade skill (0.1.3-alpha.1 -> 0.1.5-alpha.1) plus a zero-dependency seam scanner, packaged as a bundle skill and an npx CLI.

Version-locked upgrade corridor for DSH plugins: dsh-v0.1.5-alpha.1 -> dsh-v0.1.5-rc.1 (card + skill + zero-dependency scanner)

Shared zero-runtime-dependency toolkit for PerryLink DSH plugins: a pluggable Provider registry seam, fail-closed approval and adaptive session-event gates, mechanical verify scripts, shared sanitize/pricing/judge modules, and a new-plugin skeleton.

Multi-channel decision & remote-control bridge for DeepSeek Harness: pushes any workspace's approval/question cards to IM channels (WeChat iLink first) and answers them from chat, with a session console, per-channel security, and an open push service.

Automatic strong/cheap model-tier routing for DeepSeek Harness: intent-gated tier landing on the agent/request waterfall, plan-mode handoff (strong plans, cheap implements), deterministic high-risk guards on tools/pre-execute, failure escalation with TTL fallback, a /tier command, and tier status tools