fix: harden token security, add guardrails, fix 0-test loop

apps/bot/.env.example

@@ -0,0 +1,17 @@
+# Wright Telegram Bot — environment variables
+# Copy to .env and fill in values:  cp .env.example .env
+
+# Telegram bot token from @BotFather
+BOT_TOKEN=
+
+# Supabase project URL
+SUPABASE_URL=https://xndulzwwzioghxhfgzig.supabase.co
+
+# Supabase service_role key (NOT the anon key)
+SUPABASE_KEY=
+
+# GitHub personal access token (needs repo scope for PR creation)
+GITHUB_TOKEN=
+
+# Comma-separated Telegram user IDs allowed to use the bot (empty = no restriction)
+ALLOWED_TELEGRAM_USERS=

apps/bot/Dockerfile

@@ -0,0 +1,80 @@
+# =============================================================================
+# Wright Bot -- Production Multi-Stage Dockerfile
+#
+# Stages:
+#   1. deps    -- install pnpm + all workspace dependencies
+#   2. build   -- compile TypeScript (shared + bot)
+#   3. runtime -- minimal image with built artifacts
+#
+# Build context: repo root (run with `docker build -f apps/bot/Dockerfile .`)
+# =============================================================================
+
+# ---------------------------------------------------------------------------
+# Stage 1: Install dependencies
+# ---------------------------------------------------------------------------
+FROM node:22-slim AS deps
+
+RUN corepack enable && corepack prepare pnpm@9.15.0 --activate
+
+WORKDIR /app
+
+# Copy only what pnpm needs for dependency resolution (layer cache)
+COPY package.json pnpm-workspace.yaml pnpm-lock.yaml* turbo.json ./
+COPY packages/shared/package.json packages/shared/
+COPY apps/bot/package.json apps/bot/
+
+# Install all deps (including devDependencies needed for build)
+RUN pnpm install --frozen-lockfile || pnpm install
+
+# ---------------------------------------------------------------------------
+# Stage 2: Build TypeScript
+# ---------------------------------------------------------------------------
+FROM deps AS build
+
+# Copy source files
+COPY tsconfig.base.json ./
+COPY packages/shared/tsconfig.json packages/shared/
+COPY packages/shared/src/ packages/shared/src/
+COPY apps/bot/tsconfig.json apps/bot/
+COPY apps/bot/src/ apps/bot/src/
+
+# Build shared first, then bot
+RUN pnpm --filter @wright/shared build && pnpm --filter @wright/bot build
+
+# Prune devDependencies for a leaner copy into runtime.
+RUN pnpm prune --prod \
+    && mkdir -p packages/shared/node_modules apps/bot/node_modules
+
+# ---------------------------------------------------------------------------
+# Stage 3: Runtime
+# ---------------------------------------------------------------------------
+FROM node:22-slim AS runtime
+
+WORKDIR /app
+
+# Copy workspace manifests (needed for pnpm workspace resolution at runtime)
+COPY --from=build /app/package.json /app/pnpm-workspace.yaml ./
+
+# Copy shared package (built artifacts + package.json)
+COPY --from=build /app/packages/shared/package.json packages/shared/
+COPY --from=build /app/packages/shared/dist/ packages/shared/dist/
+
+# Copy bot (built artifacts + package.json)
+COPY --from=build /app/apps/bot/package.json apps/bot/
+COPY --from=build /app/apps/bot/dist/ apps/bot/dist/
+
+# Copy production node_modules from pruned build stage
+COPY --from=build /app/node_modules/ node_modules/
+COPY --from=build /app/packages/shared/node_modules/ packages/shared/node_modules/
+COPY --from=build /app/apps/bot/node_modules/ apps/bot/node_modules/
+
+# Non-root user
+RUN groupadd --gid 1001 wright \
+    && useradd --uid 1001 --gid wright --shell /bin/bash --create-home wright \
+    && chown -R wright:wright /app
+
+USER wright
+
+WORKDIR /app/apps/bot
+
+CMD ["node", "dist/index.js"]

apps/bot/fly.toml

@@ -0,0 +1,16 @@
+# Fly.io configuration for the wright Telegram bot
+# Long-polling bot — no HTTP service needed, just a persistent process
+
+app = "wright-bot"
+primary_region = "ord"
+
+[build]
+  dockerfile = "Dockerfile"
+
+[env]
+  NODE_ENV = "production"
+
+[[vm]]
+  memory = "256mb"
+  cpu_kind = "shared"
+  cpus = 1

apps/worker/.env.example

@@ -0,0 +1,24 @@
+# Wright Worker — environment variables (for local development)
+# Copy to .env and fill in values:  cp .env.example .env
+# In production (Fly.io), these are set via `fly secrets set`.
+
+# Supabase project URL
+SUPABASE_URL=https://xndulzwwzioghxhfgzig.supabase.co
+
+# Supabase service_role key (NOT the anon key)
+SUPABASE_SERVICE_ROLE_KEY=
+
+# Anthropic API key for Claude sessions
+ANTHROPIC_API_KEY=
+
+# Optional: Claude model to use (default: claude-sonnet-4-20250514)
+# CLAUDE_MODEL=claude-sonnet-4-20250514
+
+# Optional: Max turns per Claude session loop (default: 30)
+# MAX_TURNS_PER_LOOP=30
+
+# Optional: Test timeout in seconds (default: 300)
+# TEST_TIMEOUT_SECONDS=300
+
+# Optional: Workspace directory for cloned repos (default: /tmp/wright-work)
+# WORKSPACE_DIR=/tmp/wright-work

apps/worker/src/dev-loop.ts

@@ -16,8 +16,20 @@ import { detectTestRunner, detectPackageManager, installDependencies, runTests }
 import { runClaudeSession } from './claude-session.js'
 import { existsSync, rmSync, mkdirSync } from 'fs'
 
+/**
+ * Strip Telegram HTML-like formatting delimiters (e.g. `<b>`, `</b>`, `<code>`) from
+ * raw message text so they don't leak into prompts, PR bodies, or commit messages.
+ */
+function sanitizeTaskText(text: string): string {
+  return text.replace(/<\/?[^>]+>/g, '').trim()
+}
+
 export async function runDevLoop(config: DevLoopConfig): Promise<DevLoopResult> {
   const { job } = config
+  // Sanitize task text early so all downstream uses (prompts, PR body, commit messages)
+  // receive clean text without Telegram formatting delimiters.
+  job.task = sanitizeTaskText(job.task)
+
   const maxLoops = job.max_loops
   const maxBudget = job.max_budget_usd
   const maxTurns = config.maxTurnsPerLoop || DEFAULT_MAX_TURNS_PER_LOOP
@@ -195,7 +207,7 @@ export async function runDevLoop(config: DevLoopConfig): Promise<DevLoopResult>
       })
 
       const eventType =
-        lastTestResults.failed === 0 && lastTestResults.total > 0
+        lastTestResults.failed === 0 && lastTestResults.errors === 0
           ? 'test_pass'
           : 'test_fail'
       await emit(supabase, job.id, eventType, loop, {
@@ -230,15 +242,15 @@ export async function runDevLoop(config: DevLoopConfig): Promise<DevLoopResult>
 
       // 5c. Check pass
       allTestsPassed =
-        lastTestResults.failed === 0 && lastTestResults.total > 0
+        lastTestResults.failed === 0 && lastTestResults.errors === 0
     }
 
     // 6. Commit and push
     const commitMessage = allTestsPassed
       ? `feat: ${job.task.slice(0, 60)}`
       : `wip: ${job.task.slice(0, 60)} (${lastTestResults.passed}/${lastTestResults.total} tests passing)`
 
-    const commitSha = await commitAndPush(workDir, commitMessage)
+    const commitSha = await commitAndPush(workDir, commitMessage, job.github_token)
 
     // 7. Create PR
     let prUrl: string | undefined
@@ -248,6 +260,7 @@ export async function runDevLoop(config: DevLoopConfig): Promise<DevLoopResult>
         job.task.slice(0, 70),
         buildPrBody(job.task, lastTestResults, loopsCompleted, totalCost),
         job.branch,
+        job.github_token,
       )
       await emit(supabase, job.id, 'pr_created', undefined, { prUrl })
     } catch (err) {
@@ -304,12 +317,19 @@ ${workDir}
 - Dependencies are already installed
 
 ## Rules
-1. Make changes to implement the requested task
-2. After making changes, run tests to verify they pass
-3. If tests fail, fix the code (not the tests) unless the tests are clearly wrong
-4. Keep changes minimal and focused on the task
-5. Do not refactor unrelated code
-6. Do not add unnecessary dependencies`
+1. Make changes to implement the requested task.
+2. After making changes, run tests to verify they pass.
+3. If tests fail, fix the code (not the tests) unless the tests are clearly wrong.
+4. Keep changes minimal and focused ONLY on the files directly required by the task.
+5. Do not refactor unrelated code.
+6. Do not add unnecessary dependencies.
+
+## Hard constraints — NEVER violate these
+- NEVER modify test files unless the task explicitly asks for test changes.
+- NEVER modify package manifests (package.json, pyproject.toml, Cargo.toml, go.mod, setup.py, setup.cfg, etc.) unless the task explicitly requires adding or removing a dependency.
+- NEVER modify lock files (uv.lock, pnpm-lock.yaml, package-lock.json, yarn.lock, Cargo.lock, poetry.lock, etc.) under any circumstances.
+- If the task is about documentation (README, docs/, *.md), ONLY modify documentation files. Do NOT touch source code, tests, or manifests.
+- When in doubt about whether a file is in scope, leave it unchanged.`
 }
 
 function buildInitialPrompt(

apps/worker/src/github-ops.ts

@@ -7,6 +7,10 @@ export async function cloneRepo(repoUrl: string, workDir: string, token: string,
   try {
     const cloneArgs = branch ? ['--branch', branch] : []
     await git.clone(authedUrl, workDir, cloneArgs)
+    // Strip credentials from the stored remote URL so they aren't
+    // accessible via .git/config (e.g. by the Claude subprocess).
+    const repoGit = simpleGit(workDir)
+    await repoGit.remote(['set-url', 'origin', repoUrl])
   } catch (err) {
     const msg = err instanceof Error ? err.message : String(err)
     throw new Error(sanitizeGitError(msg, cleanToken))
@@ -31,7 +35,7 @@ export async function createFeatureBranch(
   return branchName
 }
 
-export async function commitAndPush(workDir: string, message: string): Promise<string> {
+export async function commitAndPush(workDir: string, message: string, githubToken?: string): Promise<string> {
   const git = simpleGit(workDir)
 
   await git.addConfig('user.email', 'wright@openadaptai.noreply.github.com')
@@ -47,28 +51,65 @@ export async function commitAndPush(workDir: string, message: string): Promise<s
 
   await git.commit(message)
 
+  // Re-inject credentials for push (they were stripped from the remote after clone).
+  const cleanToken = githubToken?.trim()
+  if (cleanToken) {
+    const remotes = await git.remote(['-v'])
+    const originMatch = (remotes || '').match(/origin\s+(https:\/\/[^\s]+)/)
+    if (originMatch) {
+      const authedUrl = originMatch[1].replace('https://', `https://x-access-token:${cleanToken}@`)
+      await git.remote(['set-url', 'origin', authedUrl])
+    }
+  }
+
   const branchStatus = await git.branch()
   const currentBranch = branchStatus.current
-  await git.push('origin', currentBranch)
+
+  try {
+    await git.push('origin', currentBranch)
+  } catch (err) {
+    const msg = err instanceof Error ? err.message : String(err)
+    throw new Error(cleanToken ? sanitizeGitError(msg, cleanToken) : msg)
+  } finally {
+    // Strip credentials from remote again after push.
+    if (cleanToken) {
+      const remotes = await git.remote(['-v'])
+      const originMatch = (remotes || '').match(/origin\s+(https:\/\/[^\s@]+@)?([^\s]+)/)
+      if (originMatch) {
+        const plainUrl = originMatch[0].replace(/x-access-token:[^@]+@/, '')
+        await git.remote(['set-url', 'origin', plainUrl.replace(/^origin\s+/, '')]).catch(() => {})
+      }
+    }
+  }
 
   const log = await git.log({ maxCount: 1 })
   return log.latest?.hash || ''
 }
 
 /**
  * Create a pull request using the gh CLI.
+ * Requires `githubToken` so the `gh` CLI can authenticate without `gh auth login`.
  */
 export async function createPullRequest(
   workDir: string,
   title: string,
   body: string,
   baseBranch: string = 'main',
+  githubToken?: string,
 ): Promise<string> {
   const { execFileSync } = await import('child_process')
+  // Minimal env — only what `gh` needs. Avoids leaking SUPABASE_SERVICE_ROLE_KEY etc.
+  const env: Record<string, string> = {
+    PATH: process.env.PATH || '/usr/local/bin:/usr/bin:/bin',
+    HOME: process.env.HOME || '/home/wright',
+  }
+  if (githubToken) {
+    env.GH_TOKEN = githubToken
+  }
   const result = execFileSync(
     'gh',
     ['pr', 'create', '--title', title, '--body', body, '--base', baseBranch],
-    { cwd: workDir, encoding: 'utf-8' },
+    { cwd: workDir, encoding: 'utf-8', env },
   )
   return result.trim() // Returns the PR URL
 }