/goals. Active Goal Registry

Persistent tier-1 active-goal tracking that steers per-session work toward stated outcomes. Adapted from a coding-agent /goal pattern. v2 adds milestones + requirements + ASCII progress visualization.

State file

SSOT-canonical. The registry lives under the goals.active key in the operational SSOT. For every CRUD op below:

• READ the current registry: python3 scripts/memory/ssot.py get goals.active
• After editing the structure, WRITE it back atomically: python3 scripts/memory/ssot.py set goals.active '<full json>' --by agent --reason "<op>"

ssot.set mirrors memory/state/active_goals.json atomically, so the Step-5 refresh (scripts/memory/update_memory_active_goals.py, which derives the MEMORY.md block FROM that file) keeps working unchanged. Do NOT hand-edit active_goals.json directly anymore; the mirror owns it.

Schema v2 (additive over v1).

Goal entry shape (full v2):

{
  "id": "g-2026-05-09-001",
  "title": "ship the content pipeline to 5 daily posts",
  "why": "build the distribution loop before the BD ramp",
  "success_criteria": ["5/day for 14 consecutive days", "engagement >= 3 replies/post avg"],
  "created": "2026-05-09T09:30:00+00:00",
  "due": null,
  "review_after": "2026-06-08",
  "status": "active",
  "sessions_touched": 0,
  "last_advanced": null,
  "tags": ["distribution", "content"],
  "notes": [],
  "milestones": [
    {"id": "m-1", "title": "ship the Critic gate", "due": "2026-05-15", "status": "done", "completed_at": "2026-05-08T...", "depends_on": [], "notes": []},
    {"id": "m-2", "title": "5/day consecutive 7d", "due": "2026-05-23", "status": "in_progress", "completed_at": null, "depends_on": ["m-1"], "notes": []},
    {"id": "m-3", "title": "engagement >= 3 replies/post avg", "due": "2026-05-30", "status": "open", "completed_at": null, "depends_on": ["m-2"], "notes": []}
  ],
  "requirements": [
    {"id": "r-1", "type": "infra", "description": "external API wired", "status": "blocked", "blocker": "API access pending", "owner": "operator", "blocked_at": "2026-05-09T...", "notes": []},
    {"id": "r-2", "type": "user-action", "description": "review the batch every week", "status": "open", "blocker": null, "owner": "operator", "blocked_at": null, "notes": []}
  ]
}

Caps (in _caps):

• max_active: 10
• review_after_default_days: 30
• drift_threshold_days: 14
• archive_after_terminal_days: 30
• milestone_stale_days: 14 (an in_progress milestone with no progress = stale)
• requirement_block_stale_days: 7 (a blocked requirement past this = escalate)

Status states:

• Goal: active / paused / done / abandoned. Terminal: done / abandoned.
• Milestone: open / in_progress / done / skipped. Terminal: done / skipped.
• Requirement: open / done / blocked. Terminal: done.

Commands

/goals (no args) → list active

Read active_goals.json. Show a table with progress bars per goal:

🎯 active goals (2/10)

g-2026-05-09-001 ship content to 5/day                    review 2026-06-08
  milestones: [█████░░░░░] 1/3 done  (m-1 ✅ / m-2 ⏳ / m-3 ◻)
  reqs:       [░░░░░░░░░░] 0/2 done  (r-1 🚧 blocked / r-2 ◻)

g-2026-05-08-002 close a client delivery                  due 2026-05-11
  milestones: [██████████] 3/3 done
  reqs:       [█████████░] 2/3 done

If empty → "no active goals. /goals new to create one."

/goals new → interactive create with auto-proposed breakdown

This is the BIG v2 flow. Steps:

1. Phase A: gather basics via conversational prompts:
   • title (1-line outcome)
   • why (1-sentence motivation)
   • success_criteria (1-3 bullets, what does done look like?)
   • due (optional, YYYY-MM-DD or "none")
   • tags (optional)
2. Phase B: AUTO-PROPOSE milestones + requirements. Inline reasoning, NOT a subagent. Steps:
   • Read the title + why + success_criteria.
   • Propose 3-5 sequential milestones that would walk from "started" to "all success_criteria met". Each milestone has: title, suggested due (back-calculated from the goal due if set, else from review_after), status=open, depends_on (chain milestones in order: m-1 has none, m-2 depends on m-1, etc).
   • Propose 3-5 requirements (skills / infra / content / user-action / external / budget / other categories). For each: type, description, suggested owner. but ALWAYS ASK the user to confirm the owner per req. Propose agent for skill/infra/content tasks, operator for user-action / external / budget tasks; the user can override.
   • Surface the proposed breakdown as a numbered list:

     proposed milestones for "{title}":
     m-1. {title} (due {date})
     m-2. {title} (due {date}, depends m-1)
     m-3. {title} (due {date}, depends m-2)
     
     proposed requirements:
     r-1. {type}: {desc}        owner? [agent/operator/external]
     r-2. {type}: {desc}        owner? [agent/operator/external]
     r-3. {type}: {desc}        owner? [agent/operator/external]
     
     reply: 'accept' / edit (e.g. 'r-1 owner operator, drop m-3, add milestone X') / 'skip breakdown' (create the goal without milestones/reqs)

3. Phase C: iterate on the user response. Apply edits. If the user says 'skip breakdown', create the goal with empty milestones/requirements arrays. they can add them later via /goals milestone add or /goals req add.
4. Phase D: persist. Generate id as g-{YYYY-MM-DD}-{NNN}. Append to state. Set defaults. Save.
5. Post-save: run python3 scripts/memory/update_memory_active_goals.py to refresh the MEMORY.md block.

Cap check: if the active count == 10 before append → error "10 active goal cap hit. pause / done / abandon one before creating another. /goals list to see them."

/goals show <gid> → full breakdown view

Single-goal deep view. Format:

🎯 g-2026-05-09-001  ACTIVE  review 2026-06-08
title: ship the content pipeline to 5 daily posts
why: build the distribution loop before the BD ramp
tags: distribution, content

success criteria:
  ✅ none yet (no milestone metrics auto-mapped. manual flip on /goals done <gid>)
  - 5/day for 14 consecutive days
  - engagement >= 3 replies/post avg

milestones [█████░░░░░] 1/3 done
  ✅ m-1  ship the Critic gate                completed 2026-05-08
  ⏳ m-2  5/day consecutive 7d                due 2026-05-23 (depends m-1)
  ◻  m-3  engagement >= 3 replies/post avg    due 2026-05-30 (depends m-2)

requirements [░░░░░░░░░░] 0/2 done
  🚧 r-1  infra: external API wired           owner operator. blocker: API access pending (blocked 0d ago)
  ◻  r-2  user-action: review the batch weekly   owner operator

session activity:
  sessions_touched: 0
  last_advanced: never
  drift: ⚠️ 14d idle threshold (currently 0d, healthy)

notes: (none)

/goals done <gid> / /goals pause <gid> / /goals abandon <gid>

Mark the goal terminal/paused. Record <status>_at: now.

If the id is not found → "no goal {id}. /goals list to see ids." If the goal has open milestones at /goals done → ask "this goal still has 2 open milestones. mark done anyway? y/n".

/goals touch <gid> → bump sessions_touched + last_advanced

Manual touch. Auto-called by /save (Step 3.7).

/goals note <gid> "<text>" → append a note to the goal

Append {"ts": now, "note": "<text>"} to notes[].

Milestone CRUD (v2)

• /goals milestone <gid> add "<title>" [--due YYYY-MM-DD] [--depends m-X]
  • Generate m-N where N = max(existing m-ids) + 1.
  • Default depends_on = the last open or in_progress milestone (so the chain extends naturally).
• /goals milestone <gid> in-progress <mid>
  • Mark in_progress. Auto-touch the parent goal.
• /goals milestone <gid> done <mid>
  • Mark done, set completed_at. If all milestones done, suggest "all milestones complete. /goals done <gid>?"
• /goals milestone <gid> skip <mid>
  • Mark skipped. Note the reason.
• /goals milestone <gid> note <mid> "<text>"

Requirement CRUD (v2)

• /goals req <gid> add <type> "<description>"
  • Generate r-N. Always ask for the owner: prompt the user [agent/operator/external].
  • status defaults to open.
• /goals req <gid> done <rid>
• /goals req <gid> block <rid> "<text>"
  • Set status=blocked, blocker=text, blocked_at=now.
• /goals req <gid> unblock <rid>
  • Set status=open, blocker=null, blocked_at=null.
• /goals req <gid> note <rid> "<text>"

ASCII Progress Bar render (v2)

10-cell bar: [██████░░░░]. Each cell represents 10% (so 0-10 cells filled). [██████████] 6/6 done → all done. [░░░░░░░░░░] 0/3 done → none.

Render formula: cells_filled = round((done / total) * 10), clamp 0-10. If total=0, show [----------] 0/0 (gray dashes).

Used in:

• /goals list (one per goal: milestones + reqs)
• /goals show <gid> (one per group)
• MEMORY.md ## 🎯 Active Goals block (compact, milestones bar only)
• Wake-up briefing section [10] (compact, milestones bar only)

Surfacing

Layer 1. MEMORY.md (ambient, ASCII bar inline)

<!-- active-goals-start --> ... <!-- active-goals-end --> markers preserved. Format v2:

## 🎯 Active Goals

- `g-2026-05-09-001` ship content to 5/day  [█████░░░░░] 1/3 . review 2026-06-08
- `g-2026-05-08-002` close a client delivery [██████████] 3/3 . due 2026-05-11

After ANY state mutation (new / done / pause / abandon / touch / milestone-* / req-*), run:

python3 scripts/memory/update_memory_active_goals.py

Idempotent. Replaces the block in-place if markers present, prepends if not.

Layer 2. Wake-up briefing (explicit)

Section [10] ACTIVE GOALS. Per goal:

🎯 g-2026-05-09-001 ship content to 5/day   [█████░░░░░] 1/3   idle 0d
   ⚠️ r-1 blocked 3d on "API access pending"

Auto-flagged inline:

• ⚠️ drift Nd if days_idle ≥ 14
• ⚠️ overdue Nd if past due
• 🚧 r-X blocked Nd on "<reason>" for each requirement blocked >0d

Integration with /save

skills/save/SKILL.md Step 3.7 runs scripts/memory/goal_touch_session.py which scans outputs/raw/**/*.md (last 24h) for frontmatter goal: g-XXX tags and bumps sessions_touched + last_advanced for matching active goals.

Drift cron (v2 extends)

scripts/memory/goal_drift_audit.py (weekly). Findings:

• drift: zero last_advanced for ≥ drift_threshold_days
• overdue: due < today
• review-due: review_after < today
• milestone-stale (v2): milestone status=in_progress, last touch ≥ milestone_stale_days
• requirement-block-stale (v2): requirement status=blocked, blocked_at ≥ requirement_block_stale_days ago

Writes to outputs/raw/agent/{date}-goal-drift.md. Chat digest only on findings.

Hard rules

• Never auto-delete. Mark terminal + archive after 30d via a dedicated script.
• Never auto-promote paused → active.
• Terminal status is sticky.
• No scoring (Bayesian/EV) in v2. Just count + status. v3 if the drift signal proves not enough.
• Always ask for the req owner per add.
• Auto-propose milestones + reqs on /goals new but ALWAYS surface for user edit before persist.

Persona scope

Unified (not persona-split). The default persona + coach read/write the same registry. The time budget is shared across personas.

Source references

• Design params: cap=10, due=optional, drift=14d, scope=unified, drift_cron ships day 1.
• v2: auto-propose breakdown, visible-only allocation with an ASCII bar, always-ask owner.
• Existing patterns lifted: the stand-down registry shape, the state/ convention, the skills/save/SKILL.md extraction discipline.