agents: take the remote-session block from Mind's generated source

The "Remote sessions" section was hand-written here and was a pass behind: it
named `No module named pytest` as the symptom of skipping the bootstrap, which
the canonical text records as a symptom that stopped appearing when the
container image moved to Python 3.12. Three review passes in a row found a copy
that had not learned what the previous pass measured.

PyAutoMind now generates that block from policy/remote_sessions.md into any repo
carrying the markers, with a drift check. Mind and Brain opted in on 2026-08-27;
this repo and its sibling were left out because the session that did it could
attach only two organs. Adding the markers here makes it four.

The per-repo halves go, deliberately — the canonical text carries no test
counts, timings or declared deps, because those are what rot, and a wrong number
in every repo's always-loaded context is worse than no number. This repo's own
dependencies stay in .claude/session-python.txt, which the hook reads at run
time.

Also regenerates the hook copy, which changed with Mind's leg 3b.

Author: Jammy2211

.claude/hooks/session-start.sh

@@ -22,7 +22,9 @@
 #      resolving PATH without this session's env (a subprocess with a scrubbed
 #      environment, a `#!/usr/bin/env python3` script) also gets 3.12;
 #   3. the uv-managed tools rebuilt on 3.12 — mypy and flake8 read the
-#      interpreter's version, so on 3.11 they judged code against 3.11 rules.
+#      interpreter's version, so on 3.11 they judged code against 3.11 rules —
+#      and then repaired, because (2) is what breaks each tool env's own
+#      `bin/python`, and a rebuild cannot fix a link it is handed.
 #
 # What it deliberately does NOT touch: the update-alternatives links under
 # /usr/bin. Scripts with a literal `#!/usr/bin/python3` shebang follow those,
@@ -344,6 +346,69 @@ retool_uv_tools() {
     done
 }
 
+# 3b. The link uv's rebuild cannot fix from inside.
+#
+# uv creates each tool env with `bin/python` as a SYMLINK to whatever `python3`
+# was at install time — here `/usr/local/bin/python3`, which leg 2 has already
+# replaced with a wrapper that `exec`s the session venv. Every tool env's python
+# then resolves its prefix to the VENV: `sys.prefix` is the venv, the tool's own
+# site-packages never reaches `sys.path`, and the console script dies with
+# `ModuleNotFoundError: No module named 'flake8'` — with flake8 sitting
+# installed two directories away.
+#
+# TWO paths reach that state, which is why this runs after leg 3 rather than
+# only on the envs leg 3 rebuilt: leg 3 rebuilds a 3.11 tool and hands the new
+# env the hijacked path, and leg 2 separately breaks every PRE-EXISTING tool env
+# that already pointed there and that leg 3 skips (`is_py312 … && continue`).
+# Running here covers both, because leg 2 runs before leg 3.
+#
+# Measured 2026-08-27, post-bootstrap: mypy, flake8, black, poetry and pyright
+# all dead this way; `ruff` survived (a native binary) and `pytest` survived
+# (leg 2b points its shim straight at the venv, which has pytest). The
+# bootstrap's `--check` called every one of them "3.12 OK", because the
+# interpreter they reach IS 3.12 — it is simply the wrong one. A session then
+# lints clean by not linting at all, and CI is the thing that finds out.
+#
+# The fix is one link: a venv's `bin/python` must resolve to a BASE interpreter,
+# never to a path this hook hijacks.
+#
+# This lived in `scripts/session_bootstrap.sh` for one pass, which is the wrong
+# home. The bootstrap is what a MULTI-repo session runs; a SINGLE-repo session
+# registers this hook and never calls the bootstrap, so it got leg 2's breakage
+# and none of the repair.
+repair_uv_tools() {
+    local tools_dir base tool link prefix
+    base="$("$VENV/bin/python" -c 'import sys, os; print(os.path.join(sys.base_prefix, "bin", "python3.12"))' 2>/dev/null)" || base=""
+    [ -x "$base" ] || base="$(command -v python3.12 2>/dev/null)" || base=""
+    [ -x "$base" ] || return 0
+    tools_dir="${PYAUTO_UV_TOOLS_DIR:-$(uv tool dir 2>/dev/null || echo "$HOME/.local/share/uv/tools")}"
+    [ -d "$tools_dir" ] || return 0
+    for tool in "$tools_dir"/*/; do
+        link="${tool}bin/python"
+        [ -L "$link" ] || continue
+        # Ask the interpreter where it thinks it lives, rather than tracing the
+        # link: `/usr/local/bin/python3` is a WRAPPER SCRIPT (a symlink there
+        # would lose the venv — leg 2's whole note), so `readlink -f` stops at
+        # the wrapper and reports nothing about the venv behind it. sys.prefix
+        # is the outcome; anything else is the mechanism.
+        prefix="$("$link" -c 'import sys; print(sys.prefix)' 2>/dev/null)" || continue
+        [ -n "$prefix" ] || continue
+        [ "$prefix" = "${tool%/}" ] && continue   # resolves to its own env: correct
+        # Non-fatal like every other leg: an unwritable tools dir is a warning,
+        # never a failed session start.
+        ln -sfn "$base" "$link" || {
+            log "WARNING: could not repoint $(basename "${tool%/}") at $base"
+            continue
+        }
+        prefix="$("$link" -c 'import sys; print(sys.prefix)' 2>/dev/null)" || prefix=""
+        if [ "$prefix" = "${tool%/}" ]; then
+            log "repointed $(basename "${tool%/}") at $base (it resolved into $VENV, not its own env)"
+        else
+            log "WARNING: $(basename "${tool%/}") still resolves to ${prefix:-nothing} — it will not run"
+        fi
+    done
+}
+
 # 4. Honest git history.
 #
 # A remote session clones shallow. `git merge-base --is-ancestor` then LIES
@@ -461,6 +526,15 @@ if [ "${PYAUTO_SESSION_DEFINE_ONLY:-}" = "1" ]; then
     return 0 2>/dev/null || exit 0
 fi
 
+# Run leg 3b on its own, without a session start. The door
+# `scripts/session_bootstrap.sh` knocks on after it has run every repo's hook —
+# a subprocess rather than a source, so this script's `set -euo pipefail` never
+# leaks into a caller that is contractually "a bootstrap, never a gate".
+if [ "${1:-}" = "--repair-uv-tools" ]; then
+    repair_uv_tools
+    exit 0
+fi
+
 ensure_full_clone
 install_workspace_settings
 
@@ -475,6 +549,7 @@ if ensure_venv; then
     ensure_repo_extras
     point_system_default
     retool_uv_tools
+    repair_uv_tools
     point_pytest_at_venv
     point_venv_scripts_at_venv
     # Every repo in the session registers this hook, so the second copy must not

AGENTS.md

@@ -46,28 +46,52 @@ framework, the <30s tick budget, how to add a check, and the hard rules
 (observer-only, colour coding, atomic state writes). Read it when changing
 Heart's own code, not by default.
 
-## Remote sessions: knock on the door first
+<!-- repos_sync:remote:begin -->
+## Remote sessions (Claude Code on web and mobile)
 
-Measured in a web/mobile container, where this file is loaded and little else is.
+Three facts, all measured in a web/mobile container, where this file is loaded
+and little else is. They ride in every organ because a session may hold any
+subset of them — and the session that needs this most is the one holding
+several, which is exactly the session no hook fires in.
 
-- **A session holding several organs registers no SessionStart hook.** Claude
-  Code reads project hooks from the project directory, which in that layout is
-  the repos' *parent*, not a repo — so none of the Python-3.12 setup runs and
-  the session uses the container's 3.11. Knock on the door yourself in the first
-  turn:
+- **Bootstrap in the first turn, unconditionally** — before the first test
+  command, not as a remedy once something looks wrong:
 
   ```
-  bash PyAutoMind/scripts/session_bootstrap.sh          # fix it
+  bash PyAutoMind/scripts/session_bootstrap.sh          # ~10s cold, ~1s warm
   bash PyAutoMind/scripts/session_bootstrap.sh --check  # report only
   ```
 
-  The symptom of skipping it: collection `ImportError`s naming `yaml`, or
-  `No module named pytest`. Both are the session resolving a pytest that is not
-  this workspace's — never a broken test module.
+  A session holding several organs registers no SessionStart hook — Claude Code
+  reads project hooks from the project directory, which in that layout is the
+  repos' *parent*, not a repo — so nothing has set this session up. It was once
+  phrased as a remedy keyed to `No module named pytest` or collection
+  `ImportError`s naming `yaml`; that symptom stopped appearing when the
+  container image moved to Python 3.12, while the environment is still wrong in
+  ways that read like a bad command rather than a stale session (`pytest -n
+  auto` → `unrecognized arguments: -n`). The bootstrap also **unshallows the
+  clones**: a remote session clones shallow, and `git merge-base --is-ancestor`
+  then answers "not an ancestor" for a commit whose ancestry is merely absent —
+  the answer the ship and close-out procedures act on when proving a branch
+  merged.
 
-- **Run the suite in parallel.** `pytest-xdist` is a base dep of that bootstrap:
-  `python3 -m pytest -q -n auto`. This repo's 641 tests are 7.6s on one
-  core and 2.7s on four.
+- **Then run the suite in parallel.** 4 cores, subprocess-heavy suites, no
+  single slow test: about 3.5x. `python3 -m pytest -q -n auto`, with
+  `pytest-xdist` supplied by the bootstrap above.
+
+- **There is no `gh`, and installing one does not help.** A remote session
+  reaches GitHub through the `mcp__github__*` tools, already scoped to the
+  session's repos. `gh` installs in two seconds and is a trap: it authenticates,
+  then 403s every repo-scoped call, because the egress proxy serves neither the
+  REST repo paths nor GraphQL beyond a pinned set of PR-review operations — a
+  binary that looks healthy and fails everything that matters. It also defeats
+  the surface probe, which keys off `gh auth status`. Read
+  `PyAutoBrain/skills/GITHUB_ACCESS.md` at the top of any run that touches
+  GitHub; it maps each `gh` operation onto its MCP tool. Spell that path from
+  the workspace root, as written: a multi-organ session is cwd'd at the repos'
+  *parent*, so a bare `skills/…` reads as a missing file rather than a missing
+  repo prefix.
+<!-- repos_sync:remote:end -->
 
 <!-- repos_sync:history:begin -->
 ## Never rewrite history