LIVE_EMERGENCE_SCAN — Hermes cron¶

Intent¶

Run Hyperlex on a fixed query pack every N hours, emit receipts + forecasts, append the receipt ledger and score log. Never invent Brier — operators settle later via settle / score-series.

Prerequisites¶

bash install.sh   # ~/.hermes/skills/hyperlex
export HERMES_SKILL_DIR="$HOME/.hermes/skills/hyperlex"
python3 "$HERMES_SKILL_DIR/scripts/hyperlex.py" check

One-shot (manual)¶

python3 "$HERMES_SKILL_DIR/scripts/hyperlex.py" scan \
  --config "$HERMES_SKILL_DIR/examples/cron/scan-queries.json" \
  --source mock \
  --receipt --forecasts --append-log \
  --json

Hermes cron job¶

Template: examples/cron/live-emergence-scan.job.json

Field	Value
name	`hyperlex-live-emergence-scan`
skills	`["hyperlex"]`
no_agent	`true` (script is the whole job)
schedule	`0 /6 * *` (every 6h)
script	see job JSON

Register (example CLI shape — adapt to your Hermes version):

hermes cron add \
  --name "hyperlex-live-emergence-scan" \
  --cron "0 */6 * * *" \
  --skill hyperlex \
  --no-agent \
  --script 'export HERMES_SKILL_DIR="$HOME/.hermes/skills/hyperlex"; python3 "$HERMES_SKILL_DIR/scripts/hyperlex.py" scan --config "$HERMES_SKILL_DIR/examples/cron/scan-queries.json" --source mock --receipt --forecasts --append-log --json'

Or paste fields from the job JSON into the Hermes UI / cron/jobs.json carefully.

Outputs¶

Artifact	Path
Receipts	`~/.hyperlex/receipts/hyperlex_*.json`
Receipt ledger	`~/.hyperlex/receipt_ledger.jsonl`
Forecast score log	`~/.hyperlex/score_log.jsonl`
Scan summary JSON	stdout (+ optional `--out`)

Operator settle loop¶

python3 "$HERMES_SKILL_DIR/scripts/hyperlex.py" list-receipts --limit 10
# settle forecasts from score log
python3 "$HERMES_SKILL_DIR/scripts/hyperlex.py" settle --forecast-id <id> --decision TRUE
python3 "$HERMES_SKILL_DIR/scripts/hyperlex.py" score-series --mean-shift --verify-chain

Network mode¶

Replace --source mock with combined or glossary only when:

network is allowed
optional deps installed if using crawl4ai
rate limits / disk cache are acceptable (~/.hyperlex/cache/)

Offline safety: HYPERLEX_OFFLINE=1 forces non-network fallbacks.

Risk-tier → schedule coupling (advisory · v0.3.7)¶

Maps hyperstition risk tiers to recommended scan cadence. Does not auto-mutate Hermes cron — operator reviews and registers.

Tier	Cron	Interval	max_queries	source	post-hooks
LOW	`0 /12 * *`	12h	3	mock	—
MODERATE	`0 /6 * *`	6h	5	mock	—
ELEVATED	`0 /2 * *`	2h	8	combined	vector-seed + archive-export
CRITICAL	`0 * * * *`	1h	12	combined	vector-seed + archive-export

# List policy
python3 "$HERMES_SKILL_DIR/scripts/hyperlex.py" risk-schedule --list-tiers

# Direct tier → write job envelope
python3 "$HERMES_SKILL_DIR/scripts/hyperlex.py" risk-schedule \
  --tier ELEVATED --schedule-out /tmp/hlx-cron

# From seed term (Phase 5 risk)
python3 "$HERMES_SKILL_DIR/scripts/hyperlex.py" risk-schedule \
  --term "agentic slop" --domain ai --schedule-out /tmp/hlx-cron

# Equivalent simulate mode
python3 "$HERMES_SKILL_DIR/scripts/hyperlex.py" simulate --mode schedule --tier CRITICAL

Example envelopes: examples/cron/risk-tier-elevated.job.json, examples/cron/live-emergence-scan.job.json (MODERATE default).

Post-scan advisory¶

Every scan summary includes scan_risk_advisory (lineage coverage → suggested next tier / cron). Re-run risk-schedule for full Phase 5 risk before changing live jobs.

Fail-closed rules¶

Open analysis keeps provenance.brier = null
Empty score series → NOT_COMPUTABLE
Cron script should exit non-zero only on hard failures (import/path); soft source failures degrade
Risk→schedule plans keep brier: null and state: proposed