ultragoal

name	ultragoal
description	In-session OMA implement+verify — invoke /oh-my-agy:ultragoal; sequential goals + .agy/ultragoal ledger HERE (CLI optional)
argument-hint	<approved plan path or goal brief>

ultragoal (OMA / in-session)

You are already in the agent session

When invoked via /oh-my-agy:ultragoal or this ultragoal skill (including as Autopilot Phase 3), treat $ARGUMENTS as the approved plan path or goal brief and implement HERE.

• Do not require oma autopilot drive / SID / CID / revision to start coding.
• Keep a durable workspace ledger under .agy/ultragoal/.
• Canonical slash: /oh-my-agy:ultragoal.

Purpose

Durable implement + verify phase (OMX $ultragoal analogue). Runs after ralplan APPROVE and before code-review.

Maps to Autopilot active phase ultragoal (legacy: executing). Host-native tools + workspace artifacts (not Claude Task / OMX .omx paths).

Persistence for a single goal may reuse ralph discipline; multi-worker slices use team explicitly only.

Use when

• Autopilot phase is ultragoal / user says ultragoal / implement the approved plan
• Approved ralplan exists and critic verdict is APPROVE
• Work needs sequential sub-goals with a durable ledger and fresh verify evidence

Do not use when

• Spec still vague → /oh-my-agy:deep-interview
• Plan not approved → /oh-my-agy:ralplan (do not implement on REVISE)
• Only parallel independent slices without multi-goal ledger → ultrawork
• Full product pipeline orchestration → follow parent /oh-my-agy:autopilot

Artifacts (workspace ledger)

Prefer workspace-safe paths under .agy/ (not .omx / .omc):

.agy/ultragoal/<slug>/brief.md          # goal brief (from ralplan / context)
.agy/ultragoal/<slug>/status.json       # active goal + progress
.agy/ultragoal/<slug>/ledger.jsonl      # append-only events
.agy/ultragoal/<slug>/progress.md       # human-readable progress

Optional machine summary for handoff:

.agy/ultragoal/<slug>/handoff.json

status.json (minimal)

{
  "schemaVersion": 1,
  "goal": "<goal>",
  "planPath": ".agy/plans/…",
  "activeSubgoalId": "SG-001",
  "subgoals": [
    {
      "id": "SG-001",
      "title": "…",
      "acceptanceCriteria": ["testable criterion"],
      "passes": false
    }
  ],
  "verifyEvidencePath": null,
  "complete": false
}

ledger.jsonl events (one JSON object per line)

Record at least: subgoal_started, files_touched, verify_ran, subgoal_passed, blocked, complete.

Related skills

Skill	Role in this phase
skills/verify/SKILL.md	Mandatory before marking any subgoal / phase complete
skills/team/SKILL.md	Optional multi-worker DAG (explicit only)
skills/ralph/SKILL.md · /oh-my-agy:ralph	Persistence loop for single-threaded stories
skills/ultrawork/SKILL.md	Parallel independent slices inside a subgoal
skills/code-review/SKILL.md · /oh-my-agy:code-review	Next phase after ultragoal evidence

Steps (in-session)

1. Load plan — read approved ralplan + spec; refuse to code if critic verdict is not APPROVE.
2. Init brief + ledger — write brief.md / status.json; decompose into ordered subgoals with testable acceptance criteria (no boilerplate “implementation complete”).
3. Pick next subgoal — highest priority with passes: false; respect dependencies.
4. Implement — smallest correct change for that subgoal only; log files in the ledger.
5. Verify — follow skills/verify/SKILL.md with fresh command output for that subgoal (and integrated suite when required).
6. Mark passes — only when all acceptance criteria are evidenced; append ledger events.
7. Loop until every subgoal passes: true.
8. Phase handoff — write handoff.json summary; prepare for /oh-my-agy:code-review.
9. Do not self-approve merge — next phase is code-review, then ultraqa.

Gates before leaving ultragoal

• All subgoals passes: true with ledger entries
• Fresh verify evidence for the full change set (verify skill)
• No silent scope reduction vs approved plan (plan revised with reason if needed)
• Team (if used) stopped / delivers integrated; worktrees not force-cleaned
• Handoff path ready for code-review

Stop / return conditions

• Hard blocker (credentials, product decision) → report; do not invent requirements
• Same verify failure 3× with no new diagnosis → escalate; consider return to ralplan via parent autopilot
• User cancel → cancel skill
• Plan hole discovered mid-implement → stop coding; return to ralplan with findings (do not ad-hoc redesign in silence)

Anti-patterns (forbidden)

• Claiming phase complete without fresh verify output
• Skipping tests or deleting failing tests to go green
• Starting without approved ralplan
• Using team without an explicit multi-worker need / manifest
• Blocking on missing CLI session before implementing
• Writing completion into OMA state as verified: true yourself — outer gates own that when used

Final checklist

• .agy/ultragoal/<slug>/ ledger exists and is up to date
• Every acceptance criterion has evidence
• verify skill satisfied for the change set
• Ready for /oh-my-agy:code-review

---

Appendix: optional oma CLI ledger

Only when the user wants outer Autopilot durability:

oma autopilot handoff --session <id> --expected-revision <n> \
  --key ultragoal --path .agy/ultragoal/<slug>/handoff.json

oma autopilot advance --session <id> --expected-revision <n> \
  --evidence <path-to-evidence.json>

oma autopilot status --session <id>

Optional team (explicit only):

oma team start --manifest <path>
oma team tick --team <id>
oma team deliver --team <id> --task <id> …
oma team stop --team <id>

Evidence JSON must prove implement work + verify ran. Do not advance on “looks good”. In-session ledger under .agy/ultragoal/ remains the primary artifact.