daily-summary

name	daily-summary
description	Post a daily henyey mainnet validator + project summary to GitHub Discussions on stellar-experimental/henyey. Intended to be invoked once per day at 09:00 ET (13:00 UTC) by cron. Reports validator state, deploys, incidents, issues activity, watch items, and tick aggregates over the last 24h.

Daily Summary

One invocation = one Discussion post on stellar-experimental/henyey.

Preconditions

1. Discussions must be enabled on the repo with a usable category. The skill resolves repository + category IDs at runtime via GraphQL — no IDs are hardcoded — so renaming or swapping the category is fine as long as one named below exists.
2. gh auth status must be valid for the runtime user.
3. The runtime user's GitHub token must have Discussions: Write scope. GitHub does not expose this on the Repository type, so the only reliable test is to attempt createDiscussion itself: surface a FORBIDDEN/Discussions: Write error as a fatal precondition failure and instruct the operator to regenerate the fine-grained PAT with the missing scope (the project's historical PAT has lacked it). Do not silently retry — the post will fail every time until the token is updated.
4. /home/tomer/data/monitor-loop.env must exist (same env as monitor-tick). The skill loads it to find MONITOR_SESSION_ID, MONITOR_ADMIN_PORT, MONITOR_RPC_PORT.

If any precondition fails, bail out with a clear ERROR and do NOT post.

Resolve repo + category

set -a
source /home/tomer/data/monitor-loop.env
set +a

# Preferred category names, in order. First match wins.
CATEGORY_CANDIDATES=("Daily reports" "Operations" "Announcements" "General")

resolved=$(gh api graphql -f query='
  query {
    repository(owner:"stellar-experimental", name:"henyey") {
      id
      hasDiscussionsEnabled
      discussionCategories(first:25) { nodes { id name } }
    }
  }')

REPO_ID=$(printf '%s' "$resolved" | python3 -c '
import sys, json
print(json.load(sys.stdin)["data"]["repository"]["id"])')

CATEGORY_ID=$(printf '%s' "$resolved" | python3 -c '
import sys, json
data = json.load(sys.stdin)["data"]["repository"]
cats = {n["name"]: n["id"] for n in data["discussionCategories"]["nodes"]}
for cand in '"${CATEGORY_CANDIDATES[@]@Q}".split():
    if cand in cats: print(cats[cand]); break
' )

If REPO_ID or CATEGORY_ID is empty, bail out — Discussions are not configured. Otherwise continue.

Compose the body

Build the markdown body in /tmp/daily-summary-body.md. All sections below are required even if a section is empty (write _(none)_).

1. Validator section

Pull live state once at the top of the run:

INFO=$(curl -s -m 5 "http://localhost:$MONITOR_ADMIN_PORT/info")
HEALTH=$(curl -s -m 5 -X POST "http://localhost:$MONITOR_RPC_PORT" \
  -H 'Content-Type: application/json' \
  -d '{"jsonrpc":"2.0","id":1,"method":"getHealth"}')
METRICS=$(curl -s -m 5 "http://localhost:$MONITOR_ADMIN_PORT/metrics")
source "$(git rev-parse --show-toplevel)/scripts/lib/monitor-decisions.sh"
PID=$(_find_session_process "$HOME/data" "/proc" "$MONITOR_SESSION_ID")
if [ -n "$PID" ]; then
  UPTIME_SEC=$(ps -o etimes= -p "$PID" | tr -d ' ')
  RSS_KB=$(ps -o rss= -p "$PID" | tr -d ' ')
else
  UPTIME_SEC=""
  RSS_KB=""
fi

Extract:

• state (Validating / Catching Up / etc.) from INFO
• latestLedger - oldestLedger from HEALTH, plus age and status
• quorum.qset.{agree,missing,fail_at} from INFO (nested under qset, not at the top level of the quorum block; the quorum block also has node and transitive children)
• stellar_scp_timing_first_to_self_externalize_seconds, stellar_scp_timing_externalized_seconds, stellar_scp_timing_nominated_seconds from METRICS — these are scalar gauges (current value), not histograms; just read the scalar
• henyey_jemalloc_fragmentation_pct (current value)
• henyey_jemalloc_allocated_bytes (heap-allocated, current)
• henyey_startup_peak_anon_rss_mb (startup/catchup peak anon RSS, MB — issue #3226/#3227; absent until startup completes). Read the attributed phase from the startup_peak_anon_rss_mb=<N> phase=<phase> log summary line.
• Last memory_report=true (or Memory report summary) line for heap_components_mb trajectory if available

Build (uptime, deploys, RSS-GB, frag-pct) and emit:

## Validator

- Build: `<sha>` (uptime <Hh Mm>; <N> deploys in last 24h, ~<X>m avg/build)
- State: <icon> <state> (`age=<s>s`, `agree=<n>/missing=<n>`, `lag_ms=<n>`)
- SCP propagation (`first_to_self_externalize_seconds`): <ms>ms avg
- SCP slot-cycle (`externalized_seconds`): <s>s avg
- Memory: RSS <G>G, frag <pct>% (<heap-components-trajectory>)
- Startup peak: <G>G (phase=<phase>) — _(omit if `henyey_startup_peak_anon_rss_mb` absent / node still starting up)_
- Disk: data <pct>% (<used>G of <total>G), session <size>G, mainnet <size>G

Icon: 🟢 Validating, 🟡 Catching Up, 🔴 anything else.

2. Deploys section

SINCE_ISO=$(date -u -d '24 hours ago' +%Y-%m-%dT%H:%M:%SZ)
git log --since="$SINCE_ISO" --pretty=format:'%h %s' origin/main

For each commit, also list the touched crates (top-level crates/<name>/ prefixes from git show --stat --format=). Cross-check against the session's monitor.log.preredeploy-* rotation timestamps to confirm the commit was actually deployed (not just landed on main between two deploy ticks). Mark a watch-sentinel column:

• silent — no warnings between this deploy and the next
• fired and recovered — at least one tick reported WARNING / ACTION but the validator stayed up
• rolled back — preredeploy rotation followed by a .crashed-* / .frozen-* rotation within an hour

## Deploys (<N>)

- `<sha>` <subject> <(linked PR if any)>
  - Files touched: <crates/...>
  - Watch sentinel: <silent | fired and recovered | rolled back>

If N=0, still print the heading and _(none)_.

3. Incidents section

An incident is any one of:

• The validator process died (a monitor.log.crashed-* or monitor.log.frozen-* rotation appeared in the last 24h).
• A urgent-labeled GH issue was filed in the last 24h.
• A MONITOR ACTION (deploy-driven restart) with an associated warning or sync-failure in the same tick.

For each:

- <icon> <one-line headline>
  - Window: <start UTC> → <end UTC> (<Hh Mm> total)
  - Symptom: <hash mismatch | wedge | OOM | …>
  - Root cause: <one sentence>
  - Resolved by: <commit-sha> (<linked PR/issue>)

Icon: 🔴 still-open or unresolved, 🟢 resolved.

4. Issues activity

Three sub-lists. Use gh issue list:

DAY=$(date -u +%Y-%m-%d)
gh issue list --repo stellar-experimental/henyey \
  --search "created:>=$(date -u -d '24 hours ago' +%Y-%m-%d)" \
  --state all --limit 50 --json number,title,labels,state

Bucket into:

• Filed today — created in last 24h: gh issue list --search "created:>=$(date -u -d '24 hours ago' +%Y-%m-%d)" --state all
• Closed today — closed in last 24h: gh issue list --search "closed:>=$(date -u -d '24 hours ago' +%Y-%m-%d)" --state closed
• Still open — long-running open issues (filed > 24h ago, currently open): gh issue list --search "is:open created:<$(date -u -d '24 hours ago' +%Y-%m-%d)" --state open

For each line, prefix with severity indicator. Precedence: if both urgent and not-ready labels are present, render as urgent (🔴).

• urgent → 🔴 (red dot)
• not-ready → ⚫ (grey dot)
• (no label) → ⚪ (white circle)

## Issues activity

Filed today (<N>):
- 🔴 #<n> <title>
- ⚪ #<n> <title>

Closed today (<N>):
- #<n> <title> (resolved by `<sha>`)

Still open (<N>):
- ⚪ #<n> <title> — last activity <date>

5. Watch items

Multi-tick non-incident concerns. Source: the watch array of yesterday's tick-history.jsonl entries. Aggregate the same key across ticks; report current value + 24h delta + linked issue if any.

Include startup_peak_mb= watch entries (emitted by monitor-tick check 7 from henyey_startup_peak_anon_rss_mb) so peak creep across restarts is visible — a peak that climbs run-over-run signals the cold-catchup transient is growing.

HIST=/home/tomer/data/$MONITOR_SESSION_ID/tick-history.jsonl
python3 - "$HIST" <<'PY'
# … aggregate and print watch lines …
PY

Format:

## Watch items

- `pruning_gap`: 2451 → 2640 (+189 in 24h), tracked by #1989
- `<other>`: <trajectory>, <linked issue>

If the array is empty for the window, print _(none)_.

6. Tick aggregates

The tick-history.jsonl schema evolved when monitor-tick was refactored to the TOML/evaluator catalog. Older rows (pre-refactor) use a top-level status field with values OK/ACTION/WARNING/OFFLINE; newer rows use tick (same values) and include an evaluator block with the alarm aggregate (metrics, ratio, recovery_stalled strings). The aggregator below reads either field and is forward-compatible.

python3 - "$HIST" <<'PY'
import sys, json, re
from collections import Counter
from datetime import datetime, timedelta, timezone

cutoff = datetime.now(timezone.utc) - timedelta(hours=24)
outcomes = Counter()
self_reflect = 0
firing_per_tick = []          # per-tick "N firing" counts from evaluator
ratio_breaches = 0
recovery_stall_events = 0
ratio_skipped = 0
total = 0

with open(sys.argv[1]) as f:
    for ln in f:
        try: row = json.loads(ln)
        except Exception: continue
        try:
            ts = datetime.fromisoformat(row["ts"].replace("Z","+00:00"))
        except Exception: continue
        if ts < cutoff: continue
        total += 1

        # Outcome: prefer new "tick" field, fall back to legacy "status"
        outcome = row.get("tick") or row.get("status") or ""
        if outcome in ("OK","ACTION","WARNING","OFFLINE","DEPLOY"):
            outcomes[outcome] += 1

        # Legacy self-reflect tracking (post-refactor rows omit this)
        if row.get("self_reflect","clean") != "clean":
            self_reflect += 1

        # Evaluator aggregate (new rows only)
        ev = row.get("evaluator", {})
        if ev:
            m = re.search(r"(\d+)/\d+ firing", ev.get("metrics",""))
            if m: firing_per_tick.append(int(m.group(1)))
            r = ev.get("ratio","")
            if "WARNING" in r or "breach" in r: ratio_breaches += 1
            if "skipped" in r: ratio_skipped += 1
            rs = ev.get("recovery_stalled","")
            if "delta=" in rs and "delta=0" not in rs:
                recovery_stall_events += 1

print(json.dumps({
    "total": total,
    "outcomes": dict(outcomes),
    "self_reflect": self_reflect,
    "firing_alarms_total": sum(firing_per_tick),
    "ticks_with_firing": sum(1 for n in firing_per_tick if n > 0),
    "ratio_breaches": ratio_breaches,
    "ratio_skipped": ratio_skipped,
    "recovery_stall_events": recovery_stall_events,
}, indent=2))
PY

## Tick aggregates (last 24h)

- Cron: `<id>` (every 20m) — <expected> scheduled, <actual> fired
- Outcomes: <N> OK / <N> ACTION / <N> WARNING / <N> OFFLINE / <N> DEPLOY
- Alarm-firings: <ticks-with-firing> ticks with ≥1 firing alarm
  (<firing-alarms-total> total alarm-firings across the window)
- metrics_ratio: <ratio-breaches> ticks reported breach, <ratio-skipped> skipped
- recovery_stalled: <events> non-zero-delta ticks
- Self-reflection events: <N> (legacy field; 0 expected on post-refactor rows)
- Skill commits today: <N> (<sha list>)

Skill commits = git log --since=24h --pretty=format:%h -- .claude/skills/ scripts/lib/eval-alarms.* .claude/skills/shared/.

Cron id = gh ... not applicable; use the active CronList id resolved once at the top of the run (call CronList and pick the entry whose prompt starts with Check the henyey mainnet monitor log).

7. Open questions

Optional. Only emit if the assistant has an open decision-point for the operator (e.g. "watch item drifting past tolerance — retune monitor or escalate?"). Otherwise omit the section entirely.

Post the discussion

TITLE="henyey mainnet daily — $(date -u +%Y-%m-%d)"
BODY=$(cat /tmp/daily-summary-body.md)

URL=$(gh api graphql -f query='
  mutation($r:ID!,$c:ID!,$t:String!,$b:String!) {
    createDiscussion(input:{repositoryId:$r, categoryId:$c, title:$t, body:$b}) {
      discussion { url }
    }
  }' \
  -F r="$REPO_ID" -F c="$CATEGORY_ID" -F t="$TITLE" -F b="$BODY" \
  --jq '.data.createDiscussion.discussion.url')

echo "Posted: $URL"

Always pass the body via -F b="$BODY" (variable substitution by the gh CLI), not heredoc-into--f — -f does not handle multi-line markdown cleanly. The -F form treats the value as a string variable for the GraphQL mutation, which preserves newlines and backticks. Same lesson as #1975 issue body formatting (commit 68c9efb2).

Output

Print the URL of the created discussion. If posting fails (GraphQL returned an error, or the network call failed), print the full error response and exit non-zero — cron will surface this as a failure.

Schedule

This skill is intended to fire once per day at 13:00 UTC (= 09:00 EDT, 08:00 EST — one-hour winter drift accepted). Schedule via:

CronCreate cron="0 13 * * *" prompt="/daily-summary" recurring=true

Do not co-schedule with monitor-tick — they are independent loops with no ordering dependency.