briefing

name	briefing
description	Generate today's Daily Note (or one for a specified date) at reports/daily/YYYY-MM-DD.md by aggregating the day's situation across journals, tasks, workspaces, knowledge, activity-log, and any plugin briefing hooks. Fully automated, no interaction. Use when the user asks for a daily briefing, situational analysis, or "what's going on today/yesterday".
gui	{"label":"/briefing","hint":"Generate today's situational analysis report","match":["reports/daily/**/*.md"],"arg":"none","order":60,"mode":"auto"}

/briefing — Daily Note Generation

Conduct all conversation with the user — and write all generated output — in the language defined by .claude/rules/personal-language.md (or the user's input language if that file is absent). Follow the language rules in full — exceptions and translation quality are defined in the Language Rules of .claude/rules/rill-core.md and the vault's personal-*.md overrides, never restated per skill. The English instructions below are for skill clarity, not for output style.

Tool references in this skill (Bash, Grep, Read, Edit, Glob, sub-agent) describe intent, not Claude-specific tool calls. Each harness should map them to its native equivalent — Claude Code uses its built-in tools as named; Codex CLI uses apply_patch / shell / its own sub-agent mechanism etc. as appropriate.

Generates a Daily Note that aggregates the day's situation by reading the knowledge/self/ snapshot (current-state.md, direction.md, interests.md, decisions.md) as the primary input and emitting a narrowed 6-section Daily Note (~2500 chars target). Uses internal data only, fully automated (no interaction). Aims for prose-quality readable reports.

Design essence (derivation, not aggregation): "where am I" awareness lives in knowledge/self/ — maintained by /pulse and /distill — so the briefing only derives today's actions from that snapshot instead of rebuilding a full situation analysis from raw files every morning. This keeps the note thin and prevents the one-file-does-six-jobs bloat that the pre-derivation briefing suffered from.

Arguments

$ARGUMENTS — one of the following:

• YYYY-MM-DD (e.g., 2026-03-11) → Generate Daily Note for the specified date
• Omitted → Generate Daily Note for today

Procedure

Phase 0: Initialization

1. If argument is YYYY-MM-DD: Use that date as target
2. If argument omitted: Use today's date as target
3. If reports/daily/YYYY-MM-DD.md already exists: Overwrite without confirmation (version history is in Git)
4. Run rill pages-pending-update --gc to sweep pending entries pointing at deleted pages (silent when nothing to clean; never fails)

Phase 1: Data Collection

Step A: Structured Data Collection (Script)

Execute the following in a shell and capture the YAML output:

bash .claude/commands/_lib/briefing-context.sh "$TARGET_DATE" "03:00"

Structured data returned by the script:

• workspaces: Count by status (active / completed / on_hold) + detailed list of active ones (id, days_old, last_modified, artifacts)
• inbox: Unprocessed count per subdirectory
• journals_in_window: Journal filenames within the activity window
• knowledge_created_in_window: knowledge/notes filenames created within the activity window
• tag_health: Top 5 tag counts + list of tags exceeding 50 files
• task_tickets: Ticket file statistics (open/waiting/overdue counts + due_soon list + all filenames)
• activity_window: "Yesterday" time range (based on day_boundary, default 03:00)

Step B: Content Collection (AI)

Collect the following data in parallel:

1. Task collection (from ticket files)
   • Use Grep for fast filtering of target tickets: Grep(pattern="^status: (open|waiting)", path="tasks/", glob="**/_task.md", output_mode="files_with_matches")
   • Read only matched files (skip done, cancelled, someday)
   • Also reference Step A's task_tickets statistics (counts, due_soon list) as supplementary data
   • Collect from each ticket: title (h1), status, due, scheduled (v3: scheduled-only tasks such as a dated follow-up qualify for 🟠 Urgent via past scheduled, must be extracted), mentions (projects/{id}), background (body opening), request
   • v3 — draft task collection (for the 🔵 Draft-candidate slot): additionally grep Grep(pattern="^status: draft", path="tasks/", glob="**/_task.md", output_mode="files_with_matches"), sort by created descending, Read the most recent 1 draft task body (title, frontmatter created / source / mentions, Goal, Background opening) — needed for deterministic 🔵 Draft-candidate card rendering. If 0 drafts, skip the slot
2. activity-log.md — v3: get entries within the past 5 days relative to TARGET_DATE (not wall-clock, so backfilled /briefing YYYY-MM-DD runs produce the correct slice). The activity window is anchored as: [TARGET_DATE - 5 days at day_boundary, TARGET_DATE at day_boundary]. Three subsets are used: (a) the activity_window subset (TARGET_DATE-1 → TARGET_DATE at day_boundary) feeds v2-style summaries / detection; (b) the recent 12h subset relative to TARGET_DATE (TARGET_DATE-12h → TARGET_DATE) feeds the v3 "Today's Flow" section; (c) the full 5-day range feeds the v3 untouched axis (a project / workspace / task with no touch in this window qualifies as untouched). Implementation: read the tail of activity-log.md; if the file is long, restrict to entries with timestamps within [TARGET_DATE - 5 days, TARGET_DATE]
3. reports/newsletter/ — Check if today's newsletter exists (for linking)
4. Previous briefing — Read the most recent reports/daily/ file (excluding today's). Skip if none exists

Detailed scans (journals, knowledge/notes/, past reports, past 2 weeks journal, pages/.pending, full workspace details) that were collected by the pre-migration briefing are now subsumed by self/current-state.md — do not duplicate them here.

Step C.0: /pulse on-demand pre-check

Before reading knowledge/self/current-state.md in Step C, check freshness so the briefing always reads a snapshot that is at most 5 minutes old (briefing-on-demand bypasses /pulse's 12h cooldown):

read .claude/state/pulse.json
if (now - last_pulse_at) > 5 min OR file missing:
  invoke /pulse with --force flag  # via the harness's Skill invocation primitive
# else: current-state.md is fresh enough, skip /pulse invocation

The invocation is synchronous (the harness's Skill tool blocks until /pulse returns). After /pulse completes, proceed to Step C. /pulse handles its own non-recursion contract (it never invokes /briefing back).

Step C: Read self/ Snapshot

Read the following self/ files. Skip a file silently if it is not present — decisions.md is a Phase 2 deliverable populated by /retrospective, so it may be empty or absent.

1. knowledge/self/current-state.md — pulse snapshot (high-velocity, updated by /pulse)
2. knowledge/self/direction.md — cross-project meta-direction (medium-velocity, monthly). v3: this is the primary signal for the important axis for Top filtering — workspaces / tasks mentioned in Active Projects here qualify directly. Items not in Active Projects can still qualify via the secondary signal constraints.md (Step C #6 below) — both paths are valid, the "ineligible" verdict only fires when neither direction.md nor constraints.md mentions the item
3. knowledge/self/interests.md — Deep Interests / Curiosity / Obligations / Career (medium-velocity, monthly)
4. knowledge/self/decisions.md — curated decision digest (3-month window, updated by /retrospective). If absent or empty, treat decision-digest content in Phase 2 as empty/skipped

v3 additional inputs for narrowing:

5. activity-log.md — read tail entries (recent 12 hours + past 5 days). Used for two purposes: (a) the "Today's Flow" section in Phase 2 (recent 12h, prose summary grouped into morning / afternoon / most recent); (b) the untouched axis for Top filtering — a workspace / task / project mentioned in direction.md Active Projects but with no activity-log touch in the past 5 days qualifies as untouched
6. knowledge/self/constraints.md — family / financial / health / life-event constraints. Used as a secondary signal for the important axis when an item is not in direction.md Active Projects but is tied to a constraints.md constraint (e.g. a tax-filing deadline tied to a financial constraint, a health checkup tied to a health constraint). Skip silently if absent or empty
7. status: draft tasks — already collected in Step B #1 (the v3 addendum reads the most recent 1 draft task body in full). Used here only as a cross-reference for the 🔵 Draft-candidate slot in Phase 2 (cap 1, most recent created, sourced from Step B's Read). If 0 matches, the slot is replaced by a second In-Progress entry or omitted entirely — the 5-count is not strict

These files are the primary input for Phase 2 generation. v3 narrowing logic:

• The Main-theme 1-line is rendered from the ## Current Main Theme section of self/direction.md (compress to 1 line, dense). Heading lookup contract: ## Current Main Theme is the canonical English structural key seeded by /onboarding; direction.md headings are lookup keys, not display strings. If the canonical heading is absent (a vault seeded before this convention may carry a localized heading), fall back to the first ## section of direction.md
• The "Today's Flow" 3 lines are rendered from activity-log.md recent 12h (group entries by time-of-day or by activity type, prose)
• The ⚠️ Today's Focus (Top, hierarchical) is filtered by the 3-condition product:
  1. important: mentioned in direction.md Active Projects, OR tied to a constraints.md constraint (the file actually read in Step C #6)
  2. untouched: no activity-log touch in past 5 days for that project / workspace / task
  3. pull: due / scheduled / explicit deadline / promise / doctrine Tie-break: explicit deadline > scheduled/due > general doctrine; longer delay first. If 0 candidates qualify, fall back to the most active project from activity-log past 12h (label as "★ Today's Focus (in progress)" instead of "⚠️")
• The 4 parallel cards are filled in this order: 1× 🟢 In Progress (most active in activity-log past 12h) + 1-2× 🟠 Urgent (due past / scheduled past / /pulse Section 6 execution-gap entries) + 0-1× 🔵 Draft candidate (most recent status: draft task). De-duplication rule: if the Top fell back to "in progress" (i.e. no important × untouched candidate) and the 🟢 parallel slot's most-active leader is the same project, the 🟢 slot picks the next-most-active project instead — never spend 2 of 5 slots on the same item
• Dependency-aware filtering (ADR-080 D80-6, depends-on/blocks recognition): a task that would otherwise qualify for Top or a 🟠 Urgent card but is blocked by an unmet depends-on is demoted. A task is blocked if its frontmatter depends-on: [tasks/foo, tasks/bar] lists at least one entry whose target task has status other than done / cancelled; broken depends-on links (target task file missing) also count as blocked. Blocked Top candidates → the next candidate takes Top, and the blocked item is listed in the "Narrowed out" section as Blocked: [{title}](../../tasks/{slug}/_task.md) — waiting on [{dep-title}](../../tasks/{dep-slug}/_task.md) (max 2 such lines; older candidate first). Blocked 🟠 candidates → omit silently (they live in /project {slug} review). A task with no depends-on field is unblocked by default (the common case), so the check is cheap: Read each candidate's frontmatter depends-on, then each referenced task's status. Reuse the same canonical algorithm as /refresh-project — both implement the "Dependency Resolution Algorithm" defined in the /project SKILL.md
• The "Narrowed out" section summarizes counts upfront (do not fold via <details> — Markdown viewer compatibility) — see Phase 2 Template
• Analytical work (contradictions, longitudinal observations) is not regenerated here — it lives in self/observations.md and the /retrospective skill instead

Step C.5: Stale Workspaces detection

After Step C completes, run a separate scan for stale active workspaces and surface the count in the Notes section:

Grep(pattern="^status: active", path="workspace/", glob="**/_workspace.md", output_mode="files_with_matches")
for each hit:
  read frontmatter, get updated (or created)
  if (now - updated) > 7 days: stale_count++

if stale_count >= 3:
  notes_extra_line = "⚠ Stalled 7+ days: {stale_count} workspaces (details in the next /retrospective)"

The check is cheap (1 Grep + frontmatter Reads of ~30 files). In Phase 2 the resulting line is appended to the Notes section (after the recommended-action nudges). If stale_count < 3, omit the line silently.

Step E: /retrospective Thursday nudge

After Step C.5, evaluate whether to surface a /retrospective weekly nudge at the top of Phase 2 (this is /retrospective's primary habit trigger — see that skill). All date inputs are derived from the briefing's target_date (Phase 0), not the wall clock, so backfilled or re-run briefings produce the nudge correct for that date and do not mutate retrospective.json for an unrelated current week.

read .claude/state/retrospective.json (if missing, treat as { last_period: null, skipped_periods: [], nudge_state: {} })
direct_completed_week = the ISO week before the one containing target_date (Mon..Sun)
period_id = "weekly-{Monday-of-direct_completed_week as YYYY-MM-DD}"

# Lazy-init nudge state for this period_id if the key is absent (first run for this week)
if period_id not in retrospective.json.nudge_state:
    retrospective.json.nudge_state[period_id] = { "nudge_count": 0, "last_nudge_at": null }

state = retrospective.json.nudge_state[period_id]
last_nudge_date = (state.last_nudge_at && date_only_of(state.last_nudge_at)) or null
same_day_rerun = (last_nudge_date == target_date)   # do not consume strikes on same-day briefing reruns

if retrospective.json.last_period == period_id:
    nudge = OFF                      # already ran for this period
elif period_id in retrospective.json.skipped_periods:
    nudge = OFF                      # 3-strike skip already recorded
elif target_date is Thursday:
    nudge = ON
    if not same_day_rerun:
        state.nudge_count += 1
        state.last_nudge_at = target_date_with_time
elif target_date is Friday/Saturday/Sunday:
    nudge = DELAYED                  # still surface, marked "delayed"
    if not same_day_rerun:
        state.nudge_count += 1
        state.last_nudge_at = target_date_with_time
else:                                # Mon/Tue/Wed
    nudge = OFF

# 3-strike auto-skip on Sunday — dedup the append since a same-day rerun would
# otherwise re-add an already-skipped period_id every time briefing runs that Sunday
if retrospective.json.nudge_state[period_id].nudge_count >= 3 and target_date is Sunday:
    if period_id not in retrospective.json.skipped_periods:
        retrospective.json.skipped_periods.append(period_id)
    nudge = OFF

# Persist updated retrospective.json (skip persistence in --dry-run mode or when target_date != today, since dated-backfill briefings should be read-only against state). Use merge-write semantics: read existing JSON, update only nudge_state[period_id] + (when applicable) skipped_periods; never touch last_run_at / last_period / pending_finalize / run_count — those are owned by the /retrospective skill.

If nudge == ON or DELAYED, prepend a single-line nudge to the Phase 2 output (above the Snapshot section). The nudge text follows DETECTED_LANG; one-line, prose tone, ending with the literal command name (do not include rill CLI commands per the Notes-section convention; /retrospective weekly is a slash command, not a shell command, so it is allowed).

State file path: .claude/state/retrospective.json. The file is initialized lazily; the first weekly run of /retrospective writes it. If the file does not exist, treat all fields as empty/null and do not error.

Failure modes:

• .claude/state/retrospective.json parse error → log a one-line warning, treat as missing, do not surface a nudge
• Today is not in a valid weekday → skip silently (defensive; date library bugs)

The check is cheap (one stat + one short JSON parse + date arithmetic). Does not require external skill invocation; if you do detect nudge_count mutation, write the updated state back to disk before exiting Phase 1.

Phase 1.5: Plugin Hook Data Collection

Collect data from plugin briefing hooks.

1. Read plugins/.enabled to get the list of enabled plugins. If the file does not exist or is empty, skip this phase
2. For each enabled plugin name, Read plugins/{name}/plugin.md frontmatter. Identify plugins with hooks.briefing field
3. If no matching plugins, skip (no message needed)
4. For each matching plugin:
   • Read the hook prompt file (plugins/{plugin-name}/{hooks.briefing path})
   • Execute via a sub-agent (general-purpose). Pass the following context:
     • Target date: The date determined in Phase 0
     • Plugin path: plugins/{plugin-name}/
   • Failures are non-fatal: Log and skip errors during hook execution
   • Collect the complete ## Section Name section (Markdown) returned by the hook
5. Insert collected hook sections in Phase 2

Phase 2: Daily Note Generation

Write the report body in the language defined by .claude/rules/personal-language.md (English when absent). Translation quality follows the Language Rules in .claude/rules/rill-core.md and the vault's personal-*.md overrides — in particular, never invent literal calques that do not exist in the target language.

File Creation

First create the file with rill mkfile (to ensure timestamp accuracy):

rill mkfile reports/daily --type daily-note --field "journal-count=N"
# For specific date: rill mkfile reports/daily --date YYYY-MM-DD --type daily-note --field "journal-count=N"

Then use Edit to append the body to the output path (frontmatter is already generated by rill mkfile).

Template (v3 — 5-item narrowing + Top hierarchy + 4-element schema, target ~2500 chars)

The Phase 2 output is structured as 6 sections after the H1 + main-theme 1-line.

Design essence (v3, canonical here): v2 surfaced every active workspace in parallel (30+ items at scale) and pushed the triage back onto the user. v3 narrows the surface to 5 items via the important × untouched × pull product, expands only the Top item into a hierarchical tree, and renders every card in the fixed 4-element schema (Stuck / Next step / Resume / Source), so each surfaced item arrives with its stuck-point, one concrete ≤30-min action, a resume command, and its sources. Not-yet-ticketed work enters only through status: draft tasks (Review-mode approval), never as free-floating suggestions.

Composition target (not strict, fall back when source-set is empty):

Slot	Count	Source / extraction
Top (⚠️ important × untouched, or ★ Today's Focus)	1	3-condition product (Step C, v3 narrowing); fall back to most-active project if 0 candidates
Parallel 🟢 In Progress	1	activity-log most-active project in past 12h
Parallel 🟠 Urgent	1-2	due past / scheduled past / /pulse Section 6 execution-gap entries
Parallel 🔵 Draft candidate	0-1	Most recent status: draft task; if 0, replace with a second In Progress or omit

Total 5 items by default; the cap is soft (3-7 is acceptable when source-set varies).

4-element schema (single-tier cards):

• Stuck (1-2 sentences, ~80 chars) — Why stuck / where it is jammed. Pulled from _workspace.md Current Position / Session History tail, or _task.md Background tail + due/scheduled delay
• Next step (1 sentence, ~60 chars) — Concrete action ≤30min to start. Pulled in this order: (a) /pulse Section 6 entry / (b) _workspace.md ## Next Steps first unchecked / (c) _task.md ## Plan next step / (d) the decision itself if it's a judgment gate
• Resume (1 backtick line) — Slash command (/focus {id} for workspace, /solve {slug} for task, link-only if judgment gate, /solve {slug} with an "(after Review-mode approval)" suffix for draft task)
• Source (1-3 Markdown links) — [name](path) to source workspace / task / journal

Top hierarchy (3-tier):

The Top item only is expanded in 3 tiers inside an ASCII tree block. Use Unicode box-drawing characters (├ └ │). Tiers are typically: Strategy (the decided means; state "not up for re-deliberation" if already decided) → Execution ← untouched (concrete action delayed N days, the stuck node) → Contingency (guardrails / fallback scenarios). If the dependency chain is shallow, collapse to 2 tiers; do not pad to 4.

# YYYY-MM-DD Daily Briefing

(SC-01 contract: H1 must match `# YYYY-MM-DD Daily Briefing` exactly. Weekday goes into the main-theme line below, not into the H1.)

(If Step E set `nudge == ON or DELAYED`, prepend ONE prose line above the main-theme line — no heading, no list bullet. Tone: warm, conversational, ending with the literal command `/retrospective weekly`. Marker word "delayed" or equivalent in `DETECTED_LANG` if nudge state is DELAYED. Omit this line entirely if `nudge == OFF`.)

**Main theme**: {the direction.md Current Main Theme section compressed to 1 line, dense — include career or other sub-directions only when direction.md itself carries them}. **Today ({weekday})**: {time constraints, if any}.

## Today's Flow (touched in the last 12 hours)

- **Morning**: {activity-log entries grouped, prose, 1 line}
- **Afternoon**: {activity-log entries grouped, prose, 1 line}
- **Most recent**: {most recent activity, prose, 1 line}

(If no activity in past 12h, write a single line: "No activity in the last 12 hours". This section is the v3 equivalent of v2's "Yesterday's Activity" — it shifts focus from "yesterday-aggregated" to "what's already happening today".)

## ⚠️ Today's Focus — important but untouched

**[{Project}] {Top item title}**

> {1-2 sentence stuck explanation: why it qualifies as important × untouched, with specific signals — direction.md mention / activity-log gap / deadline / promise / doctrine}

├ Strategy: {if already decided, say so explicitly ("not up for re-deliberation") + 1 line of what was decided} ├ Execution ← untouched: {concrete action N days delayed, the stuck node — be specific about the delay and what's blocking} └ Contingency: {guardrails / fallback scenarios — G1-Gn if applicable, otherwise 1 line naming the fallback when stalled}

**Resume** → `{/focus or /solve command}`
**Source** → [{name}]({path}) · [{name}]({path})

(If 0 candidates qualify for important × untouched, replace the heading with `## ★ Today's Focus (in progress)` and the meta line with "moving forward on today's momentum"; the tree may still apply if dependencies exist, otherwise output as a single-tier card.)

## In Parallel (4)

### 🟢 In Progress — [{Project}] {item title}

- **Stuck**: {what's happening today, what's left}
- **Next step**: {concrete next action}
- **Resume**: `{/focus or /solve command, or "this session" if currently active}`
- **Source**: [{name}]({path})

### 🟠 Urgent — [{Project}] {item title}

- **Stuck**: {expiry delay, blocker}
- **Next step**: {concrete action ≤30min}
- **Resume**: `/solve {slug}`
- **Source**: [task ticket]({path})

### 🟠 Urgent — [{Project}] {item title}

(Same 4-element structure; omit this card if only 1 Urgent item exists.)

### 🔵 Draft candidate — [{Project}] {item title}

- **Stuck**: {why this is surfacing now, why it's not yet a real task}
- **Next step**: approve the draft task in Review mode → `status: open` → run `/solve`
- **Resume**: `/solve {slug}` (after Review-mode approval)
- **Source**: [draft task]({path}) · {related rules / source files}

(Omit this card entirely if 0 `status: draft` tasks exist. The 5-count is not strict.)

## Narrowed out

(Do **not** fold via `<details>` — count-summaries go on the heading line in bold. Markdown viewer compatibility.)

**Active workspaces, remaining N** · not tied to today's decisions: X / stalled 15-29 days: Y (next week's `/close` candidates) / dormant 30+ days: Z (`/retrospective` Sunday auto-skip candidates) → details in [knowledge/self/current-state.md](../../knowledge/self/current-state.md)

**Waiting tasks N** · overdue: X / due this month: Y / one-off scheduled: Z (if any overdue, surface exactly 1 inline link)

**External signals** · today: critical alerts X, releases Y · prior day: critical alerts Z, releases W — latest in [newsletter]({path})

## Notes

(Bulleted prose, **5 items maximum**. Include:
- **Today's recommended action** — Two nudge cases, both must read as a request the user makes to Claude, never as a terminal command. **Never output `rill log`, `rill clip`, or any `rill *` CLI command in the Notes section.**
  1. **Journal capture nudge** (trigger: `journal-count` from frontmatter is 0, OR the past 3+ days have had journal-count=0): warm, brief nudge in `DETECTED_LANG` inviting the user to start capturing. Examples: *"When you're ready to capture a thought, click 'New Entry' in the Rill app, or open a Claude Code session in your vault and just tell me what's on your mind — I'll handle the rest. No commands needed."*
  2. **Inbox processing nudge** (trigger: any `inbox.*.unprocessed` count is non-zero): prose nudge stating counts per subdirectory and inviting the user to ask Claude to pull in the new entries. If this nudge has been appearing for several days in a row, suggest "you can also ask Claude to set this up as a daily automation."
- Other one-line observations
)

(※ Insert Plugin Hook sections collected in Phase 1.5 here, each as-is `## Section Name`.)

Empty cases:

Case	Behavior
Top "important × untouched" 0 candidates	Switch heading to ## ★ Today's Focus (in progress), use most-active project from activity-log past 12h
/pulse Section 6 0 entries	Fill 🟠 Urgent from due past / scheduled past only; if 0, set Urgent count to 0 and add a second 🟢 In Progress
status: draft task 0 entries	Omit 🔵 Draft-candidate card entirely (4 items output is fine)
activity-log past 12h 0 entries	"Today's Flow" section: single line "No activity in the last 12 hours"
All active workspaces 0	Do not drop Top + parallel cards wholesale. Re-source them entirely from tasks/ instead: Top = highest-pull task (overdue, then scheduled past, then due near, then status: draft); parallel = remaining Urgent tasks (overdue / scheduled past / /pulse Section 6) + draft tasks. Top still uses the same ## ⚠️ Today's Focus heading but the hierarchical tree degrades to 1-tier (no Strategy / Contingency sub-branches when the source is a single task). Add an explicit note "no active workspaces — surfacing urgent items from tasks/" at the top of the Top card

Output language (headings, labels, body): The template strings above are the canonical English forms. Render the emitted Daily Note — section headings, card labels (Stuck / Next step / Resume / Source), and body — in the language defined by .claude/rules/personal-language.md (English when absent), translating the canonical strings consistently (one rendering per concept per document). Filter at generation time, not as a post-pass, and do not import vocabulary from source material in other languages — translate it. Two stability rules: (1) the H1 # YYYY-MM-DD Daily Briefing and the literal waiting status token stay verbatim in every language (SC-01 / TK-04 contracts); (2) for cross-day consistency, reuse the heading renderings of the previous Daily Note (read in Step B #4) when its language matches the current config.

Language for Notes nudges: follow DETECTED_LANG (the user's preferred language); the journal capture / inbox processing nudges keep their warm conversational tone regardless of language.

Character count target: ~2500 chars (v2 was ~1500; v3 grows because density per card is higher, but visible items drop from 8-15 to 5).

Post-output

Display a summary (3-5 lines) and finish. Do not transition to assistant mode.

Task Display Rules

Tasks from ticket files are written in rich display format.

• Title: Use the h1 (# Title) from the ticket as-is
• Background: Summarize to 1 sentence from ticket body
• Link: Relative path ../../tasks/{slug}/_task.md
• due: Display if frontmatter due exists
• status: Display in backticks for waiting

Display example:

- **[Submit Q4 expense report](../../tasks/submit-expense-report/_task.md)** — Reimbursement deadline approaching. due: 2026-04-15
- **[Schedule design review with Jane](../../tasks/phoenix-design-review/_task.md)** `waiting` — Awaiting Jane's availability for the proposal review

Rules

• Never modify inbox/journal/ and inbox/*/ original files (read-only)
• Each section should be prose-based with context and recommended actions
• If _organized/ has a same-named file, prefer reading that one
• Create reports/daily/ directory if it doesn't exist
• When referencing tasks or knowledge files in the body, always use [display name](../../relative-path) Markdown links. Backtick-only ID references (e.g., `task-xxx`) are prohibited. Same applies in prose sections
• If some self/ files are partially present (e.g. current-state.md exists but decisions.md is absent), proceed and treat the missing files as empty sections