morning-brief

Scope-only chain: pull the latest sweep snapshot, merge with Krisp commitment cache, emit a ranked list of "things Robert hasn't closed that are older than X hours."

Steps

1. Freshness check - sweep snapshot. Read

state/log/sweep-snapshots/latest.json. Compute sweep_age_hours = (now - file.mtime) / 3600. If the file is missing OR sweep_age_hours > 4, re-invoke the sweep skill and then cp <newest-timestamped-inbox.json> state/log/sweep-snapshots/latest.json before continuing. (Note: sweepAll() writes a timestamped file but does NOT update latest.json automatically - the cp is mandatory.)

1. Freshness check - Krisp cache. Check whether

~/.claude/cache/krisp/action-items.json exists and krisp_age_hours = (now - file.mtime) / 3600.

• Missing (ENOENT): set krisp_status = "missing", skip commitment

cross-check, continue to step 3.

• Stale (krisp_age_hours > 8): set krisp_status = "stale", surface

existing action_items[] but note the age in the brief header.

• Fresh (krisp_age_hours ≤ 8): set krisp_status = "ok", use

action_items[] normally.

1. For each inbox item, mark awaiting_robert = true if Robert is not the

last sender AND (krisp_status = "missing" OR no Krisp commitment exists that already promises a reply by date > now).

1. Sort by age descending. Take top 15.
2. Emit markdown digest to state/log/briefs/<date>.md and return path.

Eval

Actor: the sweep→krisp merge + ranking logic. Auditor: dedup + staleness cross-check.

const unique_ids = new Set(items.map(i => i.channel + ":" + i.id));
const has_dupes = unique_ids.size < items.length;
const all_aged = items.every(i => i.age_hours > 0);
const sweep_stale = sweep_snapshot_age_hours > 4;   // tightened from 6h
const krisp_missing = krisp_status === "missing";
const krisp_stale   = krisp_status === "stale";     // mtime > 8h

score("morning-brief", run_id, {
  score:
    items.length === 0 && reason_empty ? 1.0 :
    items.length === 0 && !reason_empty ? 0.0 :
    has_dupes ? 0.0 :
    !all_aged ? 0.0 :
    sweep_stale ? 0.5 :
    krisp_stale ? 0.5 :
    krisp_missing ? 0.75 :   // degraded but not blocked
    1.0,
  item_count: items.length,
  unique_count: unique_ids.size,
  sweep_stale,
  krisp_status,
  reason_empty,
  primary_issue:
    items.length === 0 && !reason_empty ? "silent-empty" :
    has_dupes ? "duplicate-items" :
    !all_aged ? "zero-age-item" :
    sweep_stale ? "stale-sweep-snapshot" :
    krisp_stale ? "krisp-cache-stale" :
    krisp_missing ? "krisp-cache-missing" : null,
});

Catches: duplicates across channels (dedup regression), stale sweep data (>4h), zero-age items (timestamp bugs), silent empty, stale Krisp cache (>8h), missing Krisp cache (degraded path - brief still emits).

Gotchas

• Sweep snapshot staleness (>4h). Re-invoke sweep skill and cp the

timestamped output to latest.json before continuing. Using a stale snapshot without re-invoking scores 0.5 (primary_issue: "stale-sweep-snapshot"). The cp is mandatory - sweepAll() does NOT update latest.json automatically.

• Krisp cache missing. If ~/.claude/cache/krisp/action-items.json is

absent, skip commitment cross-check, set krisp_status = "missing", and continue. The brief still emits; score is 0.75, not a failure.

• Krisp cache stale (mtime >8h). Surface existing action_items[] but

tag krisp_status = "stale". Score is 0.5 (primary_issue: "krisp-cache-stale"). Do not block the brief.

• LinkedIn fetcher may emit ts: "Apr 11" (string, no year) - Date.parse

resolves to year 2001 (~200,000h age). Treat non-numeric ts as recent; do not block the score.

Graduation

Stays prose until the chain shape is stable. Sidecar lands when the ranking logic has been tuned twice.

Rubric

criteria:
  - name: brief_output_location
    kind: deterministic
    check: "The skill must write a markdown digest to 'state/log/briefs/<date>.md'."
  - name: sweep_freshness_enforced
    kind: deterministic
    check: "If 'state/log/sweep-snapshots/latest.json' is missing or older than 4 hours, the 'sweep' skill must be re-invoked and 'latest.json' updated from a new timestamped snapshot."
  - name: krisp_cache_handling
    kind: judge
    check: "The skill must correctly handle missing, stale (>8h), or fresh Krisp cache data according to the specification, including appropriate `krisp_status` and content surfacing."
  - name: item_ranking_and_count
    kind: judge
    check: "The skill must sort items by age descending, take the top 15, and mark items as 'awaiting_robert' based on the specified logic."

morning-brief - loader

Per-turn rules for the morning-brief skill. Full reference: state/skills/morning-brief/SKILL.md. Do not skip these.

Critical Rules

• ALWAYS check sweep-snapshot freshness FIRST - if state/log/sweep-snapshots/latest.json is missing or >4h old, RE-INVOKE the sweep skill, do not show yesterday's inbox
• ALWAYS dedup across channels - keys are channel + ":" + id. Duplicate items collapse the eval to 0.0
• If Krisp cache is empty (refresh.sh didn't run today), skip the commitment cross-check and set primary_issue: "krisp-cache-missing" - do NOT block the brief
• An empty brief MUST set reason_empty - silent empty (zero items, no reason) scores 0.0 (primary_issue: "silent-empty")
• Rank by awaiting_robert = true (Robert is not the last sender AND no Krisp commitment promises future reply); top 15 by age desc

Commands

| ui dashboard | state/skills/morning-brief/resources/ui.openui | |invoke: skill is prose - follow steps 1-5 in state/skills/morning-brief/SKILL.md |sweep snapshot: state/log/sweep-snapshots/latest.json |krisp cache: ~/.claude/cache/krisp/action-items.json (written by state/bin/krisp/refresh.sh; sibling meetings.json also written). Commitments live in action_items[] - there is no commitments.json. |output: state/log/briefs/<date>.md |eval log: state/log/evals.ndjson (skill: "morning-brief")

OpenUI Resource

• Skill-owned OpenUI Lang resource: state/skills/morning-brief/resources/ui.openui. Read it before rendering or editing this skill's generated component surface.
• Treat this resource as a first-class artifact of the skill, not a generic chat response. Improve it when the skill's user-facing output needs to become richer.
• System resources compose OpenUI primitives and inherit SnappyChat tokens. Use ui_contract: branded in SKILL.md only for deliberate platform or client visuals.

Known Pitfalls

• Zero-age items signal a timestamp bug - primary_issue: "zero-age-item" blocks the score
• Stale sweep snapshot scores 0.5 even if dedup passes (primary_issue: "stale-sweep-snapshot")
• Krisp cache missing is NOT a failure - it's a documented degraded path with primary_issue: "krisp-cache-missing" and the brief still emits
• Krisp cache stale (file exists and mtime >8h) is a separate degraded path: surface the existing commitments but tag primary_issue: "krisp-cache-stale" so the eval scores 0.5, not 1.0
• sweepAll() writes a timestamped 2026-MM-DDTHH-MM-inbox.json but does NOT update latest.json. After re-invoking sweep, cp <newest>-inbox.json latest.json or the staleness check keeps reading the old file.
• LinkedIn fetcher may emit ts: "Apr 11" (string, no year) - Date.parse resolves to year 2001, producing a ~200,000h age. Treat non-numeric ts as recent and note the quirk; do not block the score.
• LinkedIn fetcher timeout is swallowed silently by sweepAll() - errors: {} even when the page-load timed out and 0 LinkedIn items were returned. Log "linkedin-timeout-swallowed" in the brief's degraded-paths list when LinkedIn item count is 0; do not infer "no DMs" from absence.
• Skill router still resolves morning-brief and sweep to the legacy flat path state/skills/<name>.md. Real canonical is folder-shape: state/skills/<name>/SKILL.md (+ AGENTS.md loader). When the launcher says "load the skill page and follow its steps" and that path 404s, fall back to <name>/SKILL.md.

Self-Test

An agent reading this should correctly:

1. [ ] Re-invoke sweep skill when the snapshot is older than 4 hours rather than reusing it?
2. [ ] Set reason_empty when emitting a zero-item brief, to avoid silent-empty?
3. [ ] Continue without commitment-check when the Krisp cache is missing, marking primary_issue accordingly?

Self-report

If this loader fell short, append a line:

echo "[$(date -u +%FT%TZ)] morning-brief: <what was missing>" >> state/log/loader-feedback.log

<!-- kernel-ok: historical attribution for the PID-loader pattern. -->

Self-correcting loader (PID feedback)

This is the rule that makes loaders learn. Ported from snappy-kernel, where it demonstrably worked: the snappy-dom-cartographer skill mapped Skool admin UI once, every later ship-agent inherited the full map for free.

Before you finish, do two things:

1. Fix gaps (P - proportional). MANDATORY. If this loader didn't cover your case - if you had to read another file, run an undocumented command, work around a wrong selector, or discover a quirk - you MUST attempt an Edit to this AGENTS.md before you log.

• Default action: edit this file. Surgical. Add the missing line, table

row, or code example where it belongs. No restructuring.

• Only LOGGED is allowed when: the fix needs >10 lines, spans multiple

files, or requires a structural rewrite. In that case the state/regen/drain.sh queue picks it up asynchronously.

• "I didn't have time" / "it's minor" / "the next agent will figure it out"

are NOT valid reasons. Minor gaps compound - that's the whole point of this loop.

• Concrete example of a valid P-fix: discover agent-browser screenshot

takes path positionally not as --output <path>. One line in a table. Edit it in. Do not just log it.

• The goal: the next agent never has to leave the loader.

2. Log the result.

echo "[$(date -u +%FT%TZ)] morning-brief: <what was missing or fixed> [FIXED|LOGGED]" >> state/log/loader-feedback.log

• FIXED = you patched this loader inline (P-fix).
• LOGGED = too large for inline; the PostToolUse enqueue + Stop-hook drain

will rewrite the loader from scratch on next session-end.

Do not skip this. Every agent run must leave the system better than it found it. The loader is the setpoint; you are the sensor; the gap is the error signal; closing the gap is the correction.