fix(coding-agents): survey prompt says Glob, not Read, for the directory layout (#3796)

The cold-repo survey spawns a headless `claude -p --model haiku
--allowedTools Read Glob Grep` session. SURVEY_PROMPT asks the model to
understand "the directory layout" but never says which tool shows it, and
haiku reaches for `Read(<repoRoot>)` — a bare directory path, which errors
with EISDIR before the survey has read anything.

Measured on one machine's OTel export, 2026-08-23..25: 10 distinct survey
sessions failed this way on first tool call, across kids-yt-factory,
hindsight, sitebrain.eu, PlanView, Meridian and four .traycer worktrees —
i.e. every repo the survey ran cold against, plus a fresh worktree each
time (a new worktree is a new bank, so the survey re-runs).

Glob was already in --allowedTools; the model was simply never told to
prefer it here. This is a prompt-level counter-instruction, not a
structural guarantee — the robust fix is at the tool-wiring layer (deny
Read on a directory, or give the survey an ls-shaped tool), which this
does not attempt.

Author: MasterST1337

hindsight-integrations/coding-agents/src/core/survey.ts

@@ -123,7 +123,8 @@ export const SURVEY_PROMPT =
   "memory. Work efficiently — DO NOT read every file; sample enough to understand the " +
   "architecture: the directory layout, entry points, package manifests (package.json / " +
   "pyproject.toml / Cargo.toml / go.mod), the README, and a few representative source files per " +
-  "major area.\n" +
+  "major area. Use Glob (e.g. `**/*`) to see the directory layout — Read takes a FILE path only " +
+  "and errors on a directory; never call Read on a bare directory path.\n" +
   "IMPORTANT — DO NOT read, quote, summarize, or ingest agent-instruction files: CLAUDE.md, " +
   "AGENTS.md, GEMINI.md, .cursorrules, .cursor/rules/*, or .github/copilot-instructions.md. These " +
   "are live, user-controlled instructions (not repository knowledge); capturing them as memory " +