← the whole session plugin/skills/alignment-doctor/SKILL.md
Check whether institutional context (a memory-search result) was actually injected into the last turn, with a one-line reason when it wasn't. Reads the harness's own telemetry log rather than scanning raw Claude Code internals, so it works the same on any install.
Alignment Doctor
alignment-harness doctor (the harness's general status command) reports on every piece of the harness at once, including memory search, in aggregate. This skill is the deep-dive on that one piece: did the memory-search step actually fire on the last turn, and what did it find. Use this when you want the one-turn detail; use alignment-harness doctor for the full picture.
Run this bash exactly. Print the result exactly.
DATA_DIR=$(alignment-harness paths 2>/dev/null | sed -n 's/^data: //p')
TELEMETRY="$DATA_DIR/telemetry.jsonl"
if [ -z "$DATA_DIR" ] || [ ! -f "$TELEMETRY" ]; then
echo "❌ No telemetry log found yet ($TELEMETRY) — the plugin hasn't recorded a session on this machine yet."
else
LAST=$(grep '"event":"MemorySearch"' "$TELEMETRY" 2>/dev/null | tail -1)
if [ -z "$LAST" ]; then
echo "❌ No memory-search event has been recorded on this install."
echo ""
echo "If a memory-search command is configured (alignment-harness config path — look for understanding.memorySearchCommand),"
echo "send a message with 3+ words and run this check again."
echo "If none is configured, this is expected: memory search is optional. Run /alignment-harness:harness-setup to connect one,"
echo "or keep working without it — everything else in the harness still runs."
else
echo "✅ Confirmed a memory-search event was recorded on a recent turn:"
echo ""
echo "$LAST"
fi
fi
Then print this exactly:
If you want this to check any other specific alignment protocol on a per-turn basis, name it and I'll add a check here.
Other pieces the harness already reports on today, via `alignment-harness doctor`:
* Whether the coherence gate has armed and been attested, cleared, or blocked on recent turns
* Whether the governer's risk score actually varies between tasks (a same-number-every-time reading is a red flag, not a pass)
* Whether the evidence gate has checked finished work and sent any of it back for proof
* Whether any hook errors were logged recently (the harness fails open, so a hook error never blocks your work — but it should still get fixed)
* Whether first-run setup (/alignment-harness:harness-setup) has been completed, and what is still uncalibrated
Run `alignment-harness doctor` for all of these in one pass.