@cyber-dash-tech/revela 0.1.0

package/package.json

@@ -0,0 +1,65 @@
+{
+  "name": "@cyber-dash-tech/revela",
+  "version": "0.1.0",
+  "description": "OpenCode plugin that turns AI into an HTML slide deck generator",
+  "type": "module",
+  "main": "./index.ts",
+  "exports": {
+    ".": "./index.ts"
+  },
+  "files": [
+    "plugin.ts",
+    "lib/",
+    "tools/",
+    "skill/",
+    "designs/default/DESIGN.md",
+    "designs/minimal/DESIGN.md",
+    "designs/editorial-ribbon/DESIGN.md",
+    "domains/general/INDUSTRY.md",
+    "domains/deeptech-investment/INDUSTRY.md",
+    "domains/consulting/INDUSTRY.md",
+    "index.ts"
+  ],
+  "keywords": [
+    "opencode",
+    "opencode-plugin",
+    "slides",
+    "presentation",
+    "ai",
+    "html",
+    "deck"
+  ],
+  "repository": {
+    "type": "git",
+    "url": "https://github.com/cyber-dash-tech/revela.git"
+  },
+  "homepage": "https://github.com/cyber-dash-tech/revela#readme",
+  "bugs": {
+    "url": "https://github.com/cyber-dash-tech/revela/issues"
+  },
+  "license": "MIT",
+  "author": "cyber-dash-tech",
+  "engines": {
+    "bun": ">=1.0.0"
+  },
+  "scripts": {
+    "test": "bun test",
+    "typecheck": "tsc"
+  },
+  "peerDependencies": {
+    "@opencode-ai/plugin": "*"
+  },
+  "dependencies": {
+    "@xmldom/xmldom": "^0.9.9",
+    "fflate": "^0.8.2",
+    "jimp": "^1.6.1",
+    "mammoth": "^1.12.0",
+    "puppeteer-core": "^24.40.0",
+    "tslog": "^4.10.2",
+    "unpdf": "^1.4.0"
+  },
+  "devDependencies": {
+    "@types/bun": "^1.3.12",
+    "typescript": "^6.0.2"
+  }
+}

package/plugin.ts

@@ -0,0 +1,365 @@
+/**
+ * revela — Core OpenCode Plugin
+ *
+ * Architecture: enable/disable mode + single /revela command (DCP style)
+ *
+ * Responsibilities:
+ * 1. On load: seed built-in designs/domains + build initial _active-prompt.md
+ * 2. config hook: register /revela command (empty template, no .md file needed)
+ * 3. command.execute.before: route all sub-commands to lib/commands/ handlers
+ * 4. tool: expose revela-designs + revela-domains tools to LLM
+ * 5. experimental.chat.system.transform: inject three-layer prompt when enabled
+ * 6. chat.message: intercept @-referenced / pasted binary files → extract text → replace FilePart with TextPart
+ * 7. tool.execute.before: intercept read on DOCX/PPTX/XLSX → preRead()
+ * 8. tool.execute.after: intercept read on PDF/images → postRead()
+ */
+
+import type { Plugin } from "@opencode-ai/plugin"
+import { existsSync, readFileSync } from "fs"
+import { extname, basename } from "path"
+import { seedBuiltinDesigns } from "./lib/design/designs"
+import { seedBuiltinDomains } from "./lib/domain/domains"
+import { buildPrompt } from "./lib/prompt-builder"
+import { ACTIVE_PROMPT_FILE } from "./lib/config"
+import { ctx } from "./lib/ctx"
+import { preRead } from "./lib/read-hooks"
+import { postRead } from "./lib/read-hooks"
+import { extractDocx } from "./lib/read-hooks/extractors/docx"
+import { extractPptx } from "./lib/read-hooks/extractors/pptx"
+import { extractXlsx } from "./lib/read-hooks/extractors/xlsx"
+import { extractPdfText } from "./lib/read-hooks/extractors/pdf"
+import { handleHelp } from "./lib/commands/help"
+import { handleEnable } from "./lib/commands/enable"
+import { handleDisable } from "./lib/commands/disable"
+import {
+  handleDesignsList,
+  handleDesignsActivate,
+  handleDesignsAdd,
+} from "./lib/commands/designs"
+import {
+  handleDomainsList,
+  handleDomainsActivate,
+  handleDomainsAdd,
+} from "./lib/commands/domains"
+import designsTool from "./tools/designs"
+import domainsTool from "./tools/domains"
+import researchSaveTool from "./tools/research-save"
+import workspaceScanTool from "./tools/workspace-scan"
+import qaTool from "./tools/qa"
+import { RESEARCH_PROMPT, RESEARCH_AGENT_SIGNATURE } from "./lib/agents/research-prompt"
+import { runQA, formatReport } from "./lib/qa"
+import { log, childLog } from "./lib/log"
+
+// OpenCode internal agent signatures — used to skip system prompt injection
+// for built-in system agents (title, summary, compaction).
+const INTERNAL_AGENT_SIGNATURES = [
+  "You are a title generator",
+  "You are a helpful AI assistant tasked with summarizing conversations",
+  "Summarize what was done in this conversation",
+]
+
+// ── Helpers ────────────────────────────────────────────────────────────────
+
+/**
+ * Display a message in the conversation UI without triggering LLM
+ * and without polluting future context. Pattern from DCP.
+ */
+async function sendIgnoredMessage(
+  client: any,
+  sessionID: string,
+  text: string,
+): Promise<void> {
+  try {
+    await client.session.prompt({
+      path: { id: sessionID },
+      body: {
+        noReply: true,
+        parts: [{ type: "text", text, ignored: true }],
+      },
+    })
+  } catch (e) {
+    log.error("sendIgnoredMessage failed", { error: e instanceof Error ? e.message : String(e) })
+  }
+}
+
+// ── Plugin ─────────────────────────────────────────────────────────────────
+
+const server: Plugin = (async (pluginCtx) => {
+  const client = pluginCtx.client
+
+  // ── Startup: seed + build initial prompt ────────────────────────────────
+  try {
+    seedBuiltinDesigns()
+    seedBuiltinDomains()
+    buildPrompt()
+    log.info("revela initialized", { promptFile: ACTIVE_PROMPT_FILE })
+  } catch (e) {
+    log.error("startup failed — prompt may not be injected", { error: e instanceof Error ? e.message : String(e) })
+  }
+
+  return {
+    // ── Register /revela command + revela-research subagent ───────────────
+    config: async (opencodeConfig) => {
+      opencodeConfig.command ??= {}
+      opencodeConfig.command["revela"] = {
+        template: "",
+        description: "Revela — AI slide deck generator (enable/disable, manage designs & domains)",
+      }
+
+      // Register the research subagent.
+      // mode: "subagent" — not shown in Tab cycle, invoked via @revela-research or Task tool.
+      // Permissions: read-only on edit/bash; write allowed to create researches/ files.
+      // No model override — inherits from the calling primary agent.
+      opencodeConfig.agent ??= {}
+      opencodeConfig.agent["revela-research"] = {
+        description: "Revela research agent — searches and collects raw materials for presentations",
+        mode: "subagent",
+        prompt: RESEARCH_PROMPT,
+        permission: {
+          edit: "deny",
+          bash: {
+            "*": "deny",
+            "ls *": "allow",
+            "ls": "allow",
+          },
+          webfetch: "allow",
+        },
+      }
+    },
+
+    // ── Route all sub-commands to lib/commands/ handlers ──────────────────
+    "command.execute.before": async (input, output) => {
+      if (input.command !== "revela") return
+
+      const sessionID: string = input.sessionID ?? ""
+      const args = (input.arguments ?? "").trim().split(/\s+/).filter(Boolean) as string[]
+      const sub = args[0]?.toLowerCase() ?? ""
+      const param = args.slice(1).join(" ")
+
+      const send = (text: string) => sendIgnoredMessage(client, sessionID, text)
+
+      if (!sub) {
+        await handleHelp(send)
+        throw new Error("__REVELA_STATUS_HANDLED__")
+      }
+      if (sub === "enable") {
+        await handleEnable(send)
+        throw new Error("__REVELA_ENABLE_HANDLED__")
+      }
+      if (sub === "disable") {
+        await handleDisable(send)
+        throw new Error("__REVELA_DISABLE_HANDLED__")
+      }
+      if (sub === "designs" && !param) {
+        await handleDesignsList(send)
+        throw new Error("__REVELA_DESIGNS_LIST_HANDLED__")
+      }
+      if (sub === "designs" && param) {
+        await handleDesignsActivate(param, send)
+        throw new Error("__REVELA_DESIGNS_ACTIVATE_HANDLED__")
+      }
+      if (sub === "domains" && !param) {
+        await handleDomainsList(send)
+        throw new Error("__REVELA_DOMAINS_LIST_HANDLED__")
+      }
+      if (sub === "domains" && param) {
+        await handleDomainsActivate(param, send)
+        throw new Error("__REVELA_DOMAINS_ACTIVATE_HANDLED__")
+      }
+      if (sub === "designs-add") {
+        await handleDesignsAdd(param, send)
+        throw new Error("__REVELA_DESIGNS_ADD_HANDLED__")
+      }
+      if (sub === "domains-add") {
+        await handleDomainsAdd(param, send)
+        throw new Error("__REVELA_DOMAINS_ADD_HANDLED__")
+      }
+
+      await send(`**Unknown sub-command:** \`${sub}\`\nRun \`/revela\` to see available commands.`)
+      throw new Error("__REVELA_UNKNOWN_HANDLED__")
+    },
+
+    // ── LLM tools: designs, domains, research, qa ─────────────────────────
+    tool: {
+      "revela-designs": designsTool,
+      "revela-domains": domainsTool,
+      "revela-research-save": researchSaveTool,
+      "revela-workspace-scan": workspaceScanTool,
+      "revela-qa": qaTool,
+    },
+
+    // ── chat.message: intercept @-referenced / pasted binary files ────────
+    // When user uses @ or pastes a file, OpenCode injects it as a FilePart
+    // directly — the read tool is never called, so tool.execute.before/after
+    // hooks don't fire. This hook intercepts FileParts before LLM sees them.
+    //
+    // DOCX/PPTX/XLSX/PDF → extract text → replace with TextPart
+    // Images              → replace with TextPart hint (LLM can use read tool)
+    "chat.message": async (input, output) => {
+      if (!ctx.enabled) return
+
+      const IMAGE_EXTS = new Set([".jpg", ".jpeg", ".png", ".bmp", ".tiff", ".tif", ".webp", ".gif"])
+      const DOC_HANDLERS: Record<string, (buf: Buffer) => Promise<string>> = {
+        ".docx": extractDocx,
+        ".pptx": extractPptx,
+        ".xlsx": extractXlsx,
+        ".pdf": extractPdfText,
+      }
+
+      for (let i = 0; i < output.parts.length; i++) {
+        const part = output.parts[i] as any
+        if (part.type !== "file") continue
+        if (part.source?.type !== "file") continue
+
+        const filePath: string = part.source.path
+        const ext = extname(filePath).toLowerCase()
+        const name = basename(filePath)
+
+        try {
+          if (DOC_HANDLERS[ext]) {
+            const buf = readFileSync(filePath)
+            const text = await DOC_HANDLERS[ext](buf)
+            output.parts[i] = {
+              ...part,
+              type: "text",
+              text: `[Extracted from: ${name}]\n\n${text}`,
+            } as any
+          } else if (IMAGE_EXTS.has(ext)) {
+            output.parts[i] = {
+              ...part,
+              type: "text",
+              text: `[Image: ${name} — use the read tool if you need to view this image]`,
+            } as any
+          }
+        } catch (e) {
+          childLog("chat.message").warn("failed to process file", {
+            file: name,
+            error: e instanceof Error ? e.message : String(e),
+          })
+          // Keep original FilePart on failure — graceful degradation
+        }
+      }
+    },
+
+    // ── Inject three-layer prompt when enabled ─────────────────────────────
+    // Skip injection for:
+    //   1. revela-research subagent (has its own research-focused prompt)
+    //   2. OpenCode internal agents (title, summary, compaction)
+    "experimental.chat.system.transform": async (input, output) => {
+      if (!ctx.enabled) return
+      try {
+        // Detect which agent is running by fingerprinting output.system content.
+        // The plugin API does not expose agent name on this hook's input.
+        const systemText = output.system.join("\n")
+
+        // Skip revela-research subagent — it has its own research prompt.
+        // Also mark ctx so tool.execute.before can allow websearch for research agents.
+        if (systemText.includes(RESEARCH_AGENT_SIGNATURE)) {
+          ctx.isResearchAgent = true
+          return
+        }
+        ctx.isResearchAgent = false
+
+        // Skip OpenCode internal system agents (title generator, summary, compaction)
+        if (INTERNAL_AGENT_SIGNATURES.some((sig) => systemText.includes(sig))) return
+
+        const prompt = readFileSync(ACTIVE_PROMPT_FILE, "utf-8")
+        if (output.system.length > 0) {
+          output.system[output.system.length - 1] += "\n\n" + prompt
+        } else {
+          output.system.push(prompt)
+        }
+      } catch (e) {
+        log.error("failed to inject system prompt", { error: e instanceof Error ? e.message : String(e) })
+        // Surface the failure in the system prompt so the LLM and user are aware.
+        // This prevents a silent "revela enabled but not working" scenario.
+        output.system.push(
+          "\n\n[REVELA ERROR: Failed to load the slide generation prompt. " +
+          "Run /revela disable then /revela enable to reinitialize.]"
+        )
+      }
+    },
+
+    // ── Pre-read: intercept binary files before read executes ──────────────
+    // Handles DOCX/PPTX/XLSX — read tool would Effect.fail on these.
+    // Extracts text → writes temp .txt → redirects args.filePath.
+    //
+    // Also blocks websearch for the primary agent — websearch must be delegated
+    // to the revela-research subagent. Use webfetch for specific URLs instead.
+    "tool.execute.before": async (input, output) => {
+      if (!ctx.enabled) return
+
+      // ── Block websearch for primary agent ──────────────────────────────
+      if (input.tool === "websearch" && !ctx.isResearchAgent) {
+        throw new Error(
+          "[revela] websearch is not available for the primary agent. " +
+          "Delegate web research to the revela-research subagent via the Task tool — " +
+          "it searches systematically and saves structured findings for reuse across sessions. " +
+          "Use the webfetch tool if you need to read a specific URL directly.",
+        )
+      }
+
+      if (input.tool !== "read") return
+      try {
+        await preRead(output.args)
+      } catch (e) {
+        childLog("preRead").warn("extraction failed", {
+          filePath: (output.args as any)?.filePath,
+          error: e instanceof Error ? e.message : String(e),
+        })
+      }
+    },
+
+    // ── Post-read: transform PDF text + compress images ────────────────────
+    // Handles PDF and images — read tool succeeds with base64 attachment.
+    // PDF: extract text, remove base64. Images: jimp compress.
+    //
+    // Also handles: auto layout QA after writing slides/*.html
+    "tool.execute.after": async (input, output) => {
+      if (!ctx.enabled) return
+
+      // ── Post-read processing ───────────────────────────────────────────
+      if (input.tool === "read") {
+        try {
+          await postRead(input.args, output)
+        } catch (e) {
+          childLog("postRead").warn("processing failed", {
+            filePath: (input.args as any)?.filePath,
+            error: e instanceof Error ? e.message : String(e),
+          })
+        }
+        return
+      }
+
+      // ── Auto layout QA after writing slides/*.html ─────────────────────
+      if (input.tool === "write") {
+        const filePath: string = input.args?.filePath ?? ""
+        // Only trigger for HTML files inside a slides/ directory
+        if (!filePath.match(/slides\/[^/]+\.html$/)) return
+
+        try {
+          const report = await runQA(filePath)
+          // Only append QA report to tool output if there are issues
+          if (report.totalIssues > 0) {
+            const formatted = formatReport(report)
+            // Append to the write tool's output so the LLM sees it immediately
+            const existing = (output as any).result ?? ""
+            ;(output as any).result =
+              (existing ? existing + "\n\n" : "") +
+              "---\n\n**[revela layout QA]** Auto-check completed:\n\n" +
+              formatted
+          }
+        } catch (e) {
+          childLog("qa").warn("auto QA failed", {
+            filePath,
+            error: e instanceof Error ? e.message : String(e),
+          })
+          // Don't surface errors to the LLM — fail silently
+        }
+        return
+      }
+    },
+  }
+}) satisfies Plugin
+
+export default { id: "revela", server }