feat(harness): extend .env guard to pi harness + redact broad-grep leaks

src/lib/steps/harness/anthropic.ts

@@ -1,6 +1,9 @@
 import { type HookCallback, query } from '@anthropic-ai/claude-agent-sdk'
 
-import { toolInputTouchesSecret } from './secret-paths.js'
+import {
+  redactSecretGrepLines,
+  toolInputTouchesSecret,
+} from './secret-paths.js'
 import type { Harness, HarnessRunStepArgs, StepRunResult } from './types.js'
 
 // The official Seam MCP (same server the seam-plugin wires up).
@@ -40,6 +43,25 @@ const denySecretFileAccess: HookCallback = async (input) => {
   return {}
 }
 
+// The PreToolUse deny only fires when a tool *names* a secret path; a repo-wide
+// grep (path ".") still scans `.env` and returns its lines. Strip those from the
+// grep output before it reaches the model. Only the string form of the result is
+// handled — a shape the redactor can't parse is left untouched.
+const redactSecretsFromGrepOutput: HookCallback = async (input) => {
+  if (input.hook_event_name !== 'PostToolUse') return {}
+  if (input.tool_name !== 'Grep' || typeof input.tool_response !== 'string') {
+    return {}
+  }
+  const redacted = redactSecretGrepLines(input.tool_response)
+  if (redacted === input.tool_response) return {}
+  return {
+    hookSpecificOutput: {
+      hookEventName: 'PostToolUse',
+      updatedToolOutput: redacted,
+    },
+  }
+}
+
 // The control harness: drives the integration with the Claude Agent SDK.
 export const anthropicHarness: Harness = {
   name: 'anthropic',
@@ -76,9 +98,11 @@ export const anthropicHarness: Harness = {
         // reviews the result as a git diff afterward. Read/search tools and the
         // docs MCP are read-only, so nothing destructive runs unattended.
         permissionMode: 'acceptEdits',
-        // Block reads/writes of the developer's .env / secret files.
+        // Block reads/writes of the developer's .env / secret files, and strip
+        // any secret lines a broad grep still surfaces from them.
         hooks: {
           PreToolUse: [{ hooks: [denySecretFileAccess] }],
+          PostToolUse: [{ hooks: [redactSecretsFromGrepOutput] }],
         },
         mcpServers: {
           // Wired exactly like the seam-plugin: mcp-remote bridges to the hosted

src/lib/steps/harness/pi.ts

@@ -1,6 +1,7 @@
 import type { ToolDefinition } from '@earendil-works/pi-coding-agent'
 import { createJiti } from 'jiti'
 
+import { toolInputTouchesSecret } from './secret-paths.js'
 import type { Harness, HarnessRunStepArgs, StepRunResult } from './types.js'
 
 // The official Seam MCP — same server the anthropic harness and the seam-plugin
@@ -127,14 +128,14 @@ export const piHarness: Harness = {
     // Each factory returns a differently-parameterised ToolDefinition; widen to
     // the general type createAgentSession accepts (the per-tool schema variance
     // is irrelevant to the session, which treats them uniformly).
-    const customTools = [
+    const customTools = guardSecretFileTools([
       createReadToolDefinition(cwd),
       createEditToolDefinition(cwd),
       createWriteToolDefinition(cwd),
       createLsToolDefinition(cwd),
       createFindToolDefinition(cwd),
       createGrepToolDefinition(cwd),
-    ] as unknown as ToolDefinition[]
+    ] as unknown as ToolDefinition[])
 
     const { session } = await createAgentSession({
       model,
@@ -201,6 +202,47 @@ export const piHarness: Harness = {
   },
 }
 
+// pi's built-in file tools ship no secret-file guard, so wrap each one: a call
+// whose input targets a secret file (.env, …) is refused before it runs — parity
+// with the anthropic harness's PreToolUse deny. Only `execute` is overridden;
+// every other field (schema, renderers) is preserved by the spread.
+type PiToolExecute = (
+  toolCallId: string,
+  params: unknown,
+  signal: AbortSignal | undefined,
+  onUpdate: unknown,
+  ctx: unknown,
+) => Promise<unknown>
+
+function guardSecretFileTools(tools: ToolDefinition[]): ToolDefinition[] {
+  return tools.map((tool) => {
+    const runOriginal = (
+      tool as unknown as { execute: PiToolExecute }
+    ).execute.bind(tool)
+    const execute: PiToolExecute = async (
+      toolCallId,
+      params,
+      signal,
+      onUpdate,
+      ctx,
+    ) => {
+      if (toolInputTouchesSecret(params)) {
+        return {
+          content: [
+            {
+              type: 'text',
+              text: 'Reading .env / secret files is blocked. Load SEAM_API_KEY from the runtime environment instead.',
+            },
+          ],
+          details: {},
+        }
+      }
+      return runOriginal(toolCallId, params, signal, onUpdate, ctx)
+    }
+    return { ...tool, execute } as unknown as ToolDefinition
+  })
+}
+
 function readRole(message: unknown): string | undefined {
   const role = (message as { role?: unknown }).role
   return typeof role === 'string' ? role : undefined

src/lib/steps/harness/secret-paths.test.ts

@@ -1,6 +1,10 @@
 import { expect, test } from 'vitest'
 
-import { isSecretFilePath, toolInputTouchesSecret } from './secret-paths.js'
+import {
+  isSecretFilePath,
+  redactSecretGrepLines,
+  toolInputTouchesSecret,
+} from './secret-paths.js'
 
 test('isSecretFilePath: blocks .env and its variants', () => {
   expect(isSecretFilePath('.env')).toBe(true)
@@ -25,13 +29,44 @@ test('isSecretFilePath: allows ordinary files', () => {
 })
 
 test('toolInputTouchesSecret: detects the path across tool input fields', () => {
-  expect(toolInputTouchesSecret({ file_path: '.env' })).toBe(true) // Read/Edit/Write
-  expect(toolInputTouchesSecret({ path: './.env.local' })).toBe(true) // Grep/Glob/Ls
-  expect(toolInputTouchesSecret({ glob: '.env' })).toBe(true) // Grep glob
+  expect(toolInputTouchesSecret({ file_path: '.env' })).toBe(true) // SDK Read/Edit/Write
+  expect(toolInputTouchesSecret({ path: './.env.local' })).toBe(true) // pi Read / SDK Grep
+  expect(toolInputTouchesSecret({ glob: '.env' })).toBe(true) // Grep/Find glob
+  expect(toolInputTouchesSecret({ dir: 'config/.env.staging' })).toBe(true) // pi Ls
   expect(toolInputTouchesSecret({ file_path: '.env.example' })).toBe(false)
   expect(toolInputTouchesSecret({ file_path: 'src/app.ts' })).toBe(false)
 })
 
+test('redactSecretGrepLines: drops lines sourced from a secret file', () => {
+  const output = [
+    'src/config.ts:12:const url = process.env.DATABASE_URL',
+    '.env:1:DATABASE_URL=postgres://user:pw@host/db',
+    'config/.env.production:3:STRIPE_KEY=sk_live_abc',
+    'src/app.ts:5:import { Seam } from "seam"',
+  ].join('\n')
+  const redacted = redactSecretGrepLines(output)
+  expect(redacted).toBe(
+    [
+      'src/config.ts:12:const url = process.env.DATABASE_URL',
+      'src/app.ts:5:import { Seam } from "seam"',
+    ].join('\n'),
+  )
+  expect(redacted).not.toContain('sk_live_abc')
+  expect(redacted).not.toContain('postgres://')
+})
+
+test('redactSecretGrepLines: keeps .env.example lines and is a no-op when clean', () => {
+  const clean = ['src/a.ts:1:x', '.env.example:1:SEAM_API_KEY='].join('\n')
+  expect(redactSecretGrepLines(clean)).toBe(clean)
+})
+
+test('redactSecretGrepLines: handles files-with-matches (bare path) output', () => {
+  const output = ['src/config.ts', '.env', 'src/app.ts'].join('\n')
+  expect(redactSecretGrepLines(output)).toBe(
+    ['src/config.ts', 'src/app.ts'].join('\n'),
+  )
+})
+
 test('toolInputTouchesSecret: safe on non-object / empty input', () => {
   expect(toolInputTouchesSecret(null)).toBe(false)
   expect(toolInputTouchesSecret(undefined)).toBe(false)

src/lib/steps/harness/secret-paths.ts

@@ -4,14 +4,17 @@
 // model context. `.env.example` is the value-less template the wizard writes, so
 // it stays readable.
 
-// Tool-input fields that carry a filesystem path across the built-in tools:
-// Read/Edit/Write use `file_path`, Grep/Glob/Ls use `path`, Grep also takes a
-// `glob`, and notebook tools use `notebook_path`.
+// Tool-input fields that carry a filesystem path across the built-in tools of
+// both harnesses. Claude Agent SDK: Read/Edit/Write use `file_path`, Grep/Glob
+// use `path` + `glob`, notebooks use `notebook_path`. pi: Read/Grep/Find use
+// `path`, Grep/Find also take `glob`, Ls uses `dir`/`directory`.
 const PATH_BEARING_KEYS = [
   'file_path',
   'path',
   'notebook_path',
   'glob',
+  'dir',
+  'directory',
 ] as const
 
 // True for `.env` and `.env.*` at any directory depth, except `.env.example`.
@@ -33,3 +36,18 @@ export function toolInputTouchesSecret(toolInput: unknown): boolean {
   }
   return false
 }
+
+// Drop lines a grep surfaced from a secret file. The PreToolUse deny only fires
+// when a tool *names* a secret path, so a repo-wide grep (path ".") still scans
+// `.env` and returns its matching lines; this strips them from the output before
+// it reaches the model. Each grep line is prefixed with its source file
+// (`path:line:text` in content mode, or a bare `path` in files-with-matches
+// mode), so the leading token is the file to test.
+export function redactSecretGrepLines(output: string): string {
+  const lines = output.split('\n')
+  const kept = lines.filter((line) => {
+    const leadingPath = line.split(':', 1)[0] ?? line
+    return !isSecretFilePath(leadingPath)
+  })
+  return kept.length === lines.length ? output : kept.join('\n')
+}