Hooks silently exit 2 with no stderr when CLAGENTIC_LITE_HOME points at an invalid directory

[jmwest96]

Summary
session-start.sh and prompt-inject.sh (and likely the other hook shims using
the same pattern) fail with exit code 2 and completely empty stdout/stderr
when CLAGENTIC_LITE_HOME is set to a directory that doesn't contain
scripts/platform.sh. Claude Code surfaces this as "No stderr output," which
gives the user no actionable signal — the hook just silently blocks every
prompt.

Confirmed still present on current HEAD (a22e1a4, 2026-08-14), i.e. not
fixed by a normal clagentic-lite update.

Root cause
Both hooks do:
: "${CLAGENTIC_LITE_HOME:=/home/jwest/.clagentic/lite}"
. "$CLAGENTIC_LITE_HOME/scripts/platform.sh" 2>/dev/null || true

1. The := default only applies when the var is unset. It's a no-op if
   the var is set to any wrong/stale value — there's no validation that
   CLAGENTIC_LITE_HOME actually resolves to a real checkout.
2. Under POSIX sh/dash, . (source) is a special builtin. A "file not
   found" error inside a special builtin is a fatal shell error that
   terminates the script immediately. || true cannot catch it (it only
   fires on a normal nonzero exit, not a special-builtin fatal abort), and
   the preceding 2>/dev/null swallows dash's own diagnostic before it
   ever reaches Claude Code's hook runner. Net effect: exit 2, no stdout,
   no stderr, completely undiagnosable from the user's side.

This is easy to hit in practice: it's exactly what happens when a user has
a stale CLAGENTIC_LITE_HOME (or CLAGENTIC_HOME) left over from before the
env-var/path rename in fd7c3bc (lr-e84c), or any other environment drift
that leaves the var pointing at a directory that no longer exists.

Repro
env -i CLAGENTIC_LITE_HOME=/tmp/nonexistent-clagentic
/path/to/.claude/hooks/session-start.sh
echo $?
# -> 2, zero bytes on both stdout and stderr

Same result for prompt-inject.sh.

Suggested fix

• Before sourcing anything, validate that CLAGENTIC_LITE_HOME points at a
  real checkout (e.g. test -f "$CLAGENTIC_LITE_HOME/scripts/platform.sh")
  and print a clear message to stderr + exit nonzero with an actionable
  hint if not, e.g.:
  "clagentic-lite: CLAGENTIC_LITE_HOME=$CLAGENTIC_LITE_HOME has no
  scripts/platform.sh — check ~/.bashrc and ~/.config/clagentic/config for
  a stale path (run clagentic-lite doctor)."
• Avoid . "$X" 2>/dev/null || true as a "make missing-file source
  non-fatal" idiom under dash/POSIX sh generally — it does not work for
  special builtins. Guard with an explicit [ -f "$X" ] && . "$X" instead.
• doctor already has a fairly good "env var migration" check (added
  recently) that catches this exact class of drift when the user runs it
  directly — but the hooks themselves have no equivalent guard, so a
  user hits total silence before ever thinking to run doctor.

Environment

• clagentic-lite install at ~/.clagentic/lite, HEAD a22e1a4 (2026-08-14).
• Originally discovered via a stale CLAGENTIC_LITE_HOME left in ~/.bashrc
  after the CLAGENTIC_HOME -> CLAGENTIC_LITE_HOME / ~/.clagentic-lite ->
  ~/.clagentic/lite rename (fd7c3bc, lr-e84c). Reproduced independently of
  that specific cause via the env -i repro above, confirming it's a
  general hook robustness gap, not specific to that one rename.