openhermes 1.2.2

package/bootstrap.mjs

@@ -0,0 +1,255 @@
+import path from "node:path"
+import fs from "node:fs"
+import os from "node:os"
+import { fileURLToPath } from "node:url"
+
+const __dirname = path.dirname(fileURLToPath(import.meta.url))
+const HARNESS_DIR = path.resolve(__dirname, "harness")
+const RULES_DIR = path.join(HARNESS_DIR, "rules")
+const SKILLS_DIR = path.join(HARNESS_DIR, "skills")
+const CONSTITUTION_FILE = path.join(HARNESS_DIR, "constitution", "soul.md")
+const RUNTIME_FILE = path.join(HARNESS_DIR, "instructions", "RUNTIME.md")
+const TOOLS_SOURCE_DIR = path.resolve(__dirname, "lib", "tools")
+const USER_TOOLS_DIR = path.join(os.homedir(), ".config", "opencode", "tools")
+
+let _bootstrapCache = undefined
+
+function buildBootstrapContent() {
+  const constitution = fs.readFileSync(CONSTITUTION_FILE, "utf8")
+  const runtime = fs.readFileSync(RUNTIME_FILE, "utf8")
+
+  const router = `## AGENTS.md
+
+OpenHermes thin constitutional router. Full harness → \`${HARNESS_DIR}\\\`.
+
+## Soul
+
+Pragmatic. Concise. Task-oriented. Subagent-first. Inspect, then act. Smallest correct change. Verify, don't claim. Receipts over vibes. Recover by narrowing, not posturing. Skeptical — demand proof. Precision-first search: needle then broad, never reverse.
+
+## Safety
+
+Snapshot before mutation. Never delete unrelated files. Never assume \`%USERPROFILE%\\\\.config\\\\opencode\` is a git repo. Verify or roll back. **NEVER delete \`auth.json\`** (\`%USERPROFILE%\\\\.local\\\\share\\\\opencode\\\\auth.json\`).
+
+## Arsenal
+
+| Category | Items |
+|----------|-------|
+| **Native tools** | \`read\`, \`write\`, \`edit\`, \`glob\`, \`grep\`, \`bash\`, \`task\`, \`webfetch\`, \`skill\`, \`todowrite\`, \`todoread\` |
+| **MCP: openhermes-memory** | \`hm_put\`, \`hm_get\`, \`hm_list\`, \`hm_latest\`, \`hm_search\` |
+| **Memory recall cache** | \`openhermes/memory/recall/cache.json\` — read on session start, no MCP round-trip |
+| **Subagents** | \`explore\` (read-only), \`general\` (multi-step), \`architect\`, \`planner\`, \`build-error-resolver\`, \`code-reviewer\`, \`security-reviewer\`, \`e2e-runner\` |
+| **Plugins** | \`curator\` (checkpoints, mistakes, audit, compaction), \`autorecall\` (recall cache on \`session.created\`), \`skill-builder\` (complex session detection) |
+
+## Skills (available via \`skill\` tool)
+
+\`agent-browser\` \`batch-files\` \`caveman\` \`create-architectural-decision-record\` \`create-readme\` \`design-md\` \`diagnose\` \`docker-expert\` \`enhance-prompt\` \`find-skills\` \`grill-me\` \`grill-with-docs\` \`improve-codebase-architecture\` \`opencode-doctor\` \`opencode-ecc-lifecycle\` \`opencode-docs\` \`opencode-expert\` \`opencode-models\` \`opencode-recall\` \`react:components\` \`setup-matt-pocock-skills\` \`skill-creator\` \`squeez-expert\` \`stitch-design\` \`stitch-loop\` \`tailored-resume-generator\` \`taste-design\` \`tdd\` \`to-issues\` \`to-prd\` \`triage\` \`typescript-expert\` \`write-a-skill\` \`write-coding-standards-from-file\` \`zoom-out\`
+
+**OpenHermes-specific skills (discovered from harness):** \`api-design\` \`backend-patterns\` \`coding-standards\` \`e2e-testing\` \`frontend-patterns\` \`frontend-slides\` \`security-review\` \`strategic-compact\` \`tdd-workflow\` \`verification-loop\`
+
+## Delegation (Mandatory)
+
+Main context = coordination + verification only. Substantive work → subagent.
+
+| Trigger | Subagent |
+|---------|----------|
+| Multi-file implementation | \`architect\` or \`planner\` |
+| Build/TS error | \`build-error-resolver\` |
+| Code review | \`code-reviewer\` |
+| Security audit | \`security-reviewer\` |
+| E2E testing | \`e2e-runner\` |
+| Multi-file search/exploration | \`explore\` or \`general\` |
+| Any non-trivial multi-step | appropriate specialist |
+
+Never delegate trivial single-step ops. Subagent returns diff + summary + verification; inspect return only. Full ref: \`${RULES_DIR}\\\\delegation.md\`.
+
+## Memory — Gated & Precision-First
+
+- **Start**: Read recall cache first. If stale/missing → \`hm_latest\` for relevant classes.
+- **Before work**: Narrow \`hm_search\` by class, scope, keywords. Never read full indexes.
+- **Before close**: Query same-type mistakes (7 days). Match → \`code-reviewer\` or \`security-reviewer\`.
+- **On failure**: \`hm_search\` for similar incidents. Search memory before asking user.
+- **Precision ladder**: \`hm_latest\` → \`hm_search\` → \`hm_get\` → \`hm_list\` (last resort). Full index reads only for explicit audit/repair tasks.
+- **Anti-spam**: No obvious facts, no one-off prefs, no temp state, no low-risk mistakes. Supersede, don't duplicate. Full rules: \`${RULES_DIR}\\\\retrieval.md\`, \`${RULES_DIR}\\\\memory-management.md\`.
+
+## Self-Edit Authority
+
+| Auto | Conditional | Needs approval |
+|------|-------------|----------------|
+| Memory entries, mistakes, checkpoints, receipts | openhermes docs/schemas/templates/non-core rules patches | AGENTS.md core, model routing, permissions, config, protected settings |
+
+Full tiers: \`${RULES_DIR}\\\\self-heal.md\`.
+
+## Precedence
+
+1. User instruction. 2. Safety/legal/destructive guard. 3. Constitution (\`${HARNESS_DIR}\\\\constitution\\\`). 4. Project constraints. 5. Project decisions. 6. Verified guards. 7. Checkpoints. 8. Instincts. 9. Freeform notes. Full: \`${RULES_DIR}\\\\precedence.md\`.
+
+## Hygiene
+
+- Checkpoint on meaningful boundaries. Compress closed segments immediately.
+- After subagent return: verify → compress that block.
+- Compress proactively.
+- Skill candidates → \`/learn\` only if repeated pattern + \`hm_search\` confirms no dup. See \`${RULES_DIR}\\\\skills-management.md\`.
+- Audit triggers: openhermes/config change, repeated failures, session start when last audit >7 days. See \`${RULES_DIR}\\\\audit.md\`.
+
+## Escalation
+
+T0: observe → log mistake → smallest fix. T1: add prevention rule → verify. T2: diagnosis/specialist → backlog. T3: constrained safe mode. Full: \`${RULES_DIR}\\\\self-heal.md\`.
+
+## State
+
+- **Config root**: \`%USERPROFILE%\\\\.config\\\\opencode\`
+- **Auth**: \`%USERPROFILE%\\\\.local\\\\share\\\\opencode\\\\auth.json\` (NEVER delete)
+- **Forensic ledger**: \`%USERPROFILE%\\\\.local\\\\share\\\\opencode\\\\opencode.db\``
+
+  return [
+    `<OPENHERMES_BOOTSTRAP>\nOpenHermes v${getOwnVersion()} active. Harness: \`${HARNESS_DIR}\\\`. Memory at \`~/.config/opencode/openhermes/memory/\`. Rules at \`${RULES_DIR}\\\`. Skills discoverable via \`skill\` tool — use \`skill\` tool to list/load them.`,
+    `<OPENHERMES_CONSTITUTION>\n${constitution}\n</OPENHERMES_CONSTITUTION>`,
+    `<OPENHERMES_RUNTIME>\n${runtime}\n</OPENHERMES_RUNTIME>`,
+    `<OPENHERMES_ROUTER>\n${router}\n</OPENHERMES_ROUTER>`
+  ].join("\n\n")
+}
+
+function getOwnVersion() {
+  try {
+    const pkg = JSON.parse(fs.readFileSync(path.join(__dirname, "package.json"), "utf8"))
+    return pkg.version || "1.0.0"
+  } catch { return "1.0.0" }
+}
+
+function installToolFiles() {
+  try {
+    if (!fs.existsSync(TOOLS_SOURCE_DIR)) return
+    const files = fs.readdirSync(TOOLS_SOURCE_DIR).filter(f => f.endsWith(".mjs") && f !== "_memory.mjs")
+    if (!files.length) return
+    fs.mkdirSync(USER_TOOLS_DIR, { recursive: true })
+    const pkgVersion = getOwnVersion()
+    const markerPath = path.join(USER_TOOLS_DIR, ".openhermes-version")
+    let installedVersion = ""
+    try { installedVersion = fs.readFileSync(markerPath, "utf8").trim() } catch {}
+    if (installedVersion === pkgVersion) {
+      const existing = fs.readdirSync(USER_TOOLS_DIR).filter(f => f.endsWith(".mjs"))
+      const needed = [...files, "_memory.mjs"]
+      if (needed.every(f => existing.includes(f))) return
+    }
+    for (const f of ["_memory.mjs", ...files]) {
+      const src = path.join(TOOLS_SOURCE_DIR, f)
+      const dst = path.join(USER_TOOLS_DIR, f)
+      if (fs.existsSync(src)) fs.copyFileSync(src, dst)
+    }
+    fs.writeFileSync(markerPath, pkgVersion, "utf8")
+    process.stderr.write(`[openhermes-bootstrap] installed ${files.length + 1} tool files (v${pkgVersion})\n`)
+  } catch (err) {
+    process.stderr.write(`[openhermes-bootstrap] tool install error: ${err.message}\n`)
+  }
+}
+
+export const BootstrapPlugin = async ({ client, directory }) => {
+  installToolFiles()
+
+  const getContent = () => {
+    if (_bootstrapCache !== undefined) return _bootstrapCache
+    try {
+      _bootstrapCache = buildBootstrapContent()
+    } catch (err) {
+      console.error("[openhermes-bootstrap] failed to build bootstrap content:", err.message)
+      _bootstrapCache = null
+    }
+    return _bootstrapCache
+  }
+
+  return {
+    name: "openhermes-bootstrap",
+
+    config: async (config) => {
+      config.skills = config.skills || {}
+      config.skills.paths = config.skills.paths || []
+      if (!config.skills.paths.includes(SKILLS_DIR)) {
+        config.skills.paths.push(SKILLS_DIR)
+      }
+
+      config.agent = config.agent || {}
+      const PROMPTS_DIR = path.join(HARNESS_DIR, "prompts")
+      const p = (name) => `{file:${path.join(PROMPTS_DIR, name)}}`
+
+      const SUBAGENTS = {
+        "architect": {
+          description: "Software architecture specialist for system design",
+          mode: "subagent",
+          prompt: p("architect.txt"),
+          permission: { read: "allow", edit: "deny", bash: "deny" }
+        },
+        "build-error-resolver": {
+          description: "Build and TypeScript error resolution specialist",
+          mode: "subagent",
+          prompt: p("build-error-resolver.md"),
+          permission: { read: "allow", edit: "allow" }
+        },
+        "code-reviewer": {
+          description: "Expert code review specialist",
+          mode: "subagent",
+          prompt: p("code-reviewer.md"),
+          permission: { read: "allow", edit: "deny", bash: "deny", task: { explore: "allow", "*": "deny" } }
+        },
+        "e2e-runner": {
+          description: "End-to-end testing specialist using Playwright",
+          mode: "subagent",
+          prompt: p("e2e-runner.txt"),
+          permission: { read: "allow", edit: "allow" }
+        },
+        "explore": {
+          description: "Fast read-only codebase exploration agent",
+          mode: "subagent",
+          prompt: p("explore.md"),
+          permission: { read: "allow", grep: "allow", glob: "allow", list: "allow", edit: "deny", bash: "deny" }
+        },
+        "planner": {
+          description: "Expert planning specialist for complex features and refactoring",
+          mode: "subagent",
+          color: "#3B82F6",
+          prompt: p("planner.md"),
+          permission: { read: "allow", edit: "deny", bash: "deny" }
+        },
+        "security-reviewer": {
+          description: "Security vulnerability detection and remediation specialist",
+          mode: "subagent",
+          prompt: p("security-reviewer.md"),
+          permission: { read: "allow", edit: "deny", bash: "deny", task: { "*": "allow" } }
+        }
+      }
+      for (const [name, def] of Object.entries(SUBAGENTS)) {
+        if (!config.agent[name]) config.agent[name] = def
+      }
+
+      config.command = config.command || {}
+      const COMMANDS_DIR = path.join(HARNESS_DIR, "commands")
+      const ct = (file) => `{file:${path.join(COMMANDS_DIR, file)}}\n\n$ARGUMENTS`
+
+      const COMMANDS = {
+        "build-fix": { agent: "build-error-resolver", description: "Fix build and TypeScript errors", subtask: true, template: ct("build-fix.md") },
+        "code-review": { agent: "code-reviewer", description: "Review code for quality, security, and maintainability", subtask: true, template: ct("code-review.md") },
+        "plan": { agent: "planner", description: "Create a detailed implementation plan", subtask: true, template: ct("plan.md") },
+        "security": { agent: "security-reviewer", description: "Run comprehensive security review", subtask: true, template: ct("security.md") },
+        "doctor": { agent: "OpenHermes", description: "Run OpenCode OpenHermes health diagnostics", subtask: true, template: ct("doctor.md") },
+        "memory-search": { agent: "OpenHermes", description: "Search OpenHermes memory with LLM summarization", subtask: true, template: ct("memory-search.md") },
+        "learn": { agent: "OpenHermes", description: "Create a new skill from recent work patterns", subtask: true, template: ct("learn.md") }
+      }
+      for (const [name, def] of Object.entries(COMMANDS)) {
+        if (!config.command[name]) config.command[name] = def
+      }
+    },
+
+    "experimental.chat.messages.transform": async (_input, output) => {
+      try {
+        const bootstrap = getContent()
+        if (!bootstrap || !output.messages || !output.messages.length) return
+        const firstUser = output.messages.find(m => m && m.info && m.info.role === "user")
+        if (!firstUser || !firstUser.parts || !firstUser.parts.length) return
+        if (firstUser.parts.some(p => p.type === "text" && p.text.includes("OPENHERMES_BOOTSTRAP"))) return
+        const ref = firstUser.parts[0]
+        firstUser.parts.unshift({ ...ref, type: "text", text: bootstrap })
+      } catch (err) {
+        console.error("[openhermes-bootstrap] transform error:", err.message)
+      }
+    }
+  }
+}