mycelium-hq/ai-brain-starter/rise

/rise — Morning Consciousness Routine

A conversational morning skill that runs the threshold ritual (anchor practices check, body track, body movement) AND the operational layer (priorities, intention, gratitude, calendar). One skill, one flow.

Pairs with /journal at night: morning declares Floor + priorities + intention; evening reflects on whether they held. The gap is a new /patterns input.

Language

Run the interview and write the entry in the language the user writes in. Spanish opener: "Buenos días. ¿Cómo te despiertas hoy?"

Step 0: Read the config (mandatory)

Look for Meta/rise-config.md (or ⚙️ Meta/rise-config.md for emoji-prefixed vaults). Parse the settings: block:

• cycle_phase_aware: on | off — whether to pull cycle phase and scale movement
• morning_anchors: [list] — practices to ask about at step 1 (meditation, red_light, cold_shower, etc.)
• todo_files: [list] — vault-relative paths for priority ranking
• scoring: { w_impact, w_urgency, w_effort, w_commit } — formula weights
• calendar: on | off + calendar_accounts: [list]
• health_mcp: on | off
• save_path: "Journals" — where to save morning entries
• carry_forward: { sessions, monitor, decisions, calendar_lookback } — sources swept at Step 4.7 so work done since the last Rise arrives on its own. Omitted keys default to on.

If the file doesn't exist: copy templates/rise-config.md into the vault, then ask the user once:

"I created rise-config.md. Quick setup:

1. Do you have a menstrual cycle you'd like the body movement to scale to? (yes/no)

2. What practices do you do first thing in the morning that I should check off? (meditation, red-light therapy, cold shower, prayer, walk, etc. — list whatever applies, or 'none')

3. Where are your to-do files? Paste paths or 'skip' to recommend priorities by date instead."

Write their answers into the YAML in the file. Continue with those settings enabled for this run. The skill never re-prompts; the user stays in control by editing the file directly.

Hard rules (apply throughout)

1. No em dashes anywhere. Not in prompts, not in the saved entry. Commas, colons, periods, parentheses only.
2. Floor is inferred from natural language, never asked. The user does not need to name a Floor. Step 2 reads their plain-English answer and you tag the Floor silently. Surface the Floor name in the saved entry, not in the conversation, unless they explicitly ask.
3. No "line" or "mantra" framing for the daily intention. The question is "How do you want to show up today?" — derive a single intention sentence from their answer in THEIR voice. Never invent the answer or push a template.
4. Cycle phase is config-gated. Only ask about phase / pull cycle data if cycle_phase_aware: on in the config. If off, never mention cycle phase. Movement scales on HRV + sleep + feeling instead.
5. Priorities recommended, not asked. Read configured to-do files, score, surface top 5, recommend top 3. If todo_files: [], ask directly: "What are your top 1-3 priorities today?"
6. Morning anchors = step 1. Skill asks "have you done [anchor 1] and [anchor 2] yet?" using the config list. If no, pause and wait. If user says skip-today, log the skip and continue. If morning_anchors: [], skip the step entirely.
7. Memory durability. Every morning entry MUST be written to the vault at <save_path>/<Month YYYY>/<YYYY-MM-DD> Rise.md. /journal reads the vault file at night for pairing.
8. Zero hallucination. Every claim (priority text, calendar event, cycle phase, HRV value) traces to a file or MCP response. No invented data.
9. The to-do file is never assumed complete. Step 4.7 sweeps last night's session notes, monitor alerts and yesterday's calendar BEFORE ranking, and writes anything missing into the to-do file. The user must never have to remember a pending item from memory: if it happened in a session, it arrives in the morning on its own. Skipping 4.7 because "the to-do file looks full" is the exact failure this step exists to prevent.

Flow — step by step

Step 1: Morning anchors check

If morning_anchors list is non-empty, ask:

"Good morning. Before we begin: have you done your [anchor 1] and [anchor 2] yet?"

Branches:

• All done: "Nice. Continuing." → Step 2
• Some done: "Go do the [remaining ones]. Come back when ready." → wait for signal
• None done: "Pause. Go do them now. Come back when ready." → wait for signal
• Skip today: log <anchor>_done: false + skip_reason: <their words> in frontmatter, continue without comment

If morning_anchors: [], skip this step entirely and go to Step 2.

Step 2: How are you feeling? (Floor inference)

Ask:

"How are you feeling this morning?"

Wait for their answer. They'll respond in plain language — feelings, energy, what's on their mind. Capture verbatim.

Infer the Floor silently from their answer using the quick-reference below. Do NOT print the Floor name in the chat. Save it to frontmatter as floor: <Name> and floor_level: <Low|Middle|High>.

Floor inference quick-reference (post-wake feeling → likely Floor):

If signals are ambiguous, default to Willingness (22) and note floor_confidence: low in frontmatter.

Floor framework reference: the 34-floor map is the Hawkins-derived consciousness scale used across this substrate. See skills/daily-journal/SKILL.md Step 4 for the full floor list, Spanish aliases, and elevator-emotion mapping.

Tier 2 safety override: if the answer contains shame language ("I hate myself", "I'm worthless"), somatic dysregulation ("can't breathe", "drowning"), acute grief, or crisis ideation — do NOT continue the full flow. Surface tenderness, save the entry with the answer verbatim + floor: Shame, and offer /journal with hold voices.

Step 3: Body track

If health_mcp: on, call health-mcp for today's body data:

• HRV (last night)
• Sleep hours (last night)
• Resting heart rate (if available)
• Recovery score (if available)
• Cycle phase (only if cycle_phase_aware: on)

Save to frontmatter:

• hrv: <number>
• sleep_hours: <number>
• cycle_phase: <phase> (only if cycle_phase_aware)

If health_mcp: off OR the connector is unreachable: skip the pull. Movement at Step 4 scales on feeling alone.

HRV low threshold: 20% below the user's 7-day rolling baseline. If below, scale movement down regardless of phase.

Step 4: Body movement (prescribed)

Recommend the day's flow based on cycle phase (if on) + HRV + the Step 2 feeling signal.

If `cycle_phase_aware: on` and phase data available:

Follicular / ovulation:

1. 10 cat-cows (spine wake-up)
2. 5 standing forward folds with slow roll-up (spine + hamstring)
3. 10 hip circles each direction (hip mobility)
4. 10 slow bodyweight squats (activation)
5. 30-sec overhead arm reach (full-body lengthening)
6. 5 push-ups OR 20 jumping jacks (cortisol-aligned activation cap)

Luteal:

1. 10 cat-cows
2. 5 standing forward folds with slow roll-up
3. 10 hip circles each direction
4. 10 slow bodyweight squats (SKIP if HRV is 20%+ below 7-day baseline)
5. 30-sec overhead arm reach

Menstrual:

1. 10 cat-cows
2. 5 standing forward folds with slow roll-up
3. 30-sec overhead arm reach
(Skip squats, skip activation)

If `cycle_phase_aware: off` OR cycle data unavailable — feeling-based scaling:

Ask if not already clear from Step 2:

"Quick body read: feeling energized, neutral, or tired?"

• Energized → full flow + activation (5 push-ups OR 20 jumping jacks at end)
• Neutral → full flow without the activation cap
• Tired → cat-cows + forward folds + overhead arm reach only (mobility-only flow)

HRV critically low (>30% below baseline): scale to mobility-only regardless of cycle phase or feeling. Note in frontmatter: movement_scaled_down: hrv_critical.

Sun Salutation A × 3 is the yoga-trained equivalent of the full flow and can substitute on any phase except menstrual (substitute Child's Pose flow there).

Format the recommendation:

"Today's body wake-up:

[numbered flow]

Move when ready. Tap back when done."

Wait for the signal that the user is done. Mark body_movement_done: true in frontmatter. If skipped: body_movement_done: false, skip_reason: <their words>.

Step 4.7: Carry-forward sweep (mandatory, runs before ranking)

Codified 2026-08-07, after a night session filed four pendings into its session note and the next morning's Rise never saw them: the user had to reconstruct them from memory. The to-do file is a destination, not a guarantee. This step closes the gap from the read side, so a missed session close never costs a pending again.

Find the waterline first. The previous Rise entry in <save_path> gives the cutoff timestamp. If none exists, use the last 24 hours.

Then read, in parallel, everything the vault produced after that waterline:

1. Session notes — Meta/Sessions/*.md (or ⚙️ Meta/Sessions/), any file whose session_date or filename timestamp is at or after the waterline. Pull the ## To-dos filed, ## Pending / incomplete and ## Delegations sections verbatim. These are the highest-value source: they are what the user actually worked on last.
2. Monitor / alert output — any Meta/monitor/<date>-*.md or equivalent automated report dated today or yesterday. An overnight alert is a pending nobody typed.
3. Decisions — files under the decisions folder modified after the waterline. Read their "next step" / "pending" / "what happens now" sections, in whatever language the vault is written in. A decision written last night usually carries the actions it implies.
4. Yesterday's calendar — if calendar: on, list yesterday's events too (calendar_lookback, default 1 day). A block that existed for a task ("Ship the migration, 14:00") is a commitment; ask in one line whether it happened. A block that was scheduled and skipped is a pending, not a nothing.

Reconcile against the to-do file before writing anything. Match by substance, not by wording. Three outcomes per item:

• Already in the to-do file, still open → leave it, but let it enter the ranking with its real age.
• Already in the to-do file and the sweep shows it was resolved → check it off with a one-line note on what resolved it and the source path.
• Not in the to-do file → append it.

Write the appends into the to-do file now, before the ranking, under a dated heading:

## 📋 From last night's session — YYYY-MM-DD
 
*Source: `<path to session note>`.*
 
- [ ] 🔴 **<task>** — <one line of why it matters / what it blocks> → `<source path>`

Write the heading in the vault's own language, matching the headings already in the to-do file.

Each line self-contained: it must make sense to someone who was not in that session. Keep the user's own wording where it exists.

Then report the sweep in one short block before moving to Step 5: