megaplan
Run the canonical Megaplan planning pipeline from its Arnold plugin home.
Source facts
- Repository
- authrain-cloud-abdullahformuli/arnold
- Last source activity
- October 28, 2026 at 02:52
- Detected SKILL.md language
- English
- Stars
- 0
- Forks
- 0
Install options
The review-first prompt is selected by default. You can switch to a direct command or download a local copy.
Review the source files
Read SKILL.md and any companion files shown by SkillsMP before deciding whether to install.
File Explorer
100 filesShowing SKILL.md
SKILL.md
Source instructions · Read-only preview- name
- megaplan
- description
- Run the canonical Megaplan planning pipeline from its Arnold plugin home.
> **Authority status (M11):** This skill document is retained as zero-authority
> history. All repair, audit, and deployment authority has been migrated to
> canonical delegation. Content must not be used to materialize commands,
> grant authority, or drive automated repair actions.
# Megaplan
Use this pipeline for high-rigor planning workflows that need prep, plan,
critique, gate, revise, finalize, execute, review, and tiebreaker stages.
For epic-sized, migration, public-contract, or otherwise drift-sensitive work,
capture durable end-state intent with a North Star anchor: pass
`--north-star PATH` to `megaplan init` for a standalone plan, or declare
`anchors.north_star` in `chain.yaml` for epics.
---
## North Star Actions (v1)
### Where questions come from
North Star questions are **explicitly authored** — they are not extracted from
code, diffs, or execution traces. In v1 the pipeline reads them from a
`NORTHSTAR.md` file in the plan directory (standalone plans) and/or from
sprint-brief anchors declared in `chain.yaml` (epic chains). There is no
semantic-parity extraction step: if a question is not written in one of those
sources it does not exist from the pipeline's point of view.
### How actions flow through the pipeline
1. **Gate (structured carry).** When a gate or reviewer evaluator raises a
concern tied to a North Star question, it emits a `north_star_actions[]`
entry — an action with a concrete `action_type` (`change_plan`,
`add_gate`, `add_scenario`, `add_checker`, `dead_delete`, or
`add_human_halt`), a `severity` (blocking/advisory), supporting evidence,
and optional plan references. Actions in the *dangerous* categories
(`route_authority`, `baselines`, `row_carrier_exemptions`,
`target_narrowing`, `generated_conformance_authority`,
`live_plan_topology_resume_risk`) are **blocking by schema rule** regardless
of what the evaluator labels them. All actions are normalized at gate time
and carried forward through `gate_carry.json` (with `gate.json` as a
fallback).
2. **Revise (change the plan or halt).** The revise worker prompt receives
the carried `north_star_actions[]` with per-action-type instructions:
- `change_plan` — make a concrete, traceable plan change.
- `add_gate` — add an explicit gate requirement.
- `add_scenario` — add a concrete scenario / test case.
- `add_checker` — add an automated checker.
- `dead_delete` — remove dead plan steps.
- `add_human_halt` — halt; the action cannot be mapped to a plan change.
Before the revise worker runs, a pre-worker guard halts through the
existing `CliError`/step-failure path when any carried blocking action is
`add_human_halt`, has an unrecognized `action_type`, or is a
dangerous-category blocker with no concrete target (`plan_refs` or
`required_change`). **Revise must change the plan or halt** — it cannot
silently ignore a blocking action.
After the revise worker completes, the pipeline validates that every
carried blocking action has a corresponding `north_star_actions_addressed[]`
entry with concrete `plan_refs` and the matching `action_type` marker.
Prose-only resolutions, mismatched action types, and omitted actions are
rejected (fail-closed).
3. **Finalize / Review → done (closeout blocking).** Finalize refuses to
produce executable tasks if any carried blocking North Star action is
unresolved in the latest revise metadata. The review→done transition
runs a separate pre-check before `TransitionPolicy.evaluate_review_done()`;
unresolved blocking actions deny the transition through the existing
denial path. Absent or malformed `north_star_actions_addressed[]` is
treated as all-carried-blocking-unresolved (fail-closed).
### `north_star_critical` and robustness
Chain and milestone specs support an optional `north_star_critical: bool`
(default `false`). When `true`, the milestone (or chain) declares that
North Star enforcement is mandatory for that work unit and cannot be
skipped.
`north_star_critical: true` is **incompatible** with `bare` and `light`
robustness. The chain-spec validator rejects these combinations with a
`CliError` — it never silently upgrades the robustness level. Acceptable
robustness levels are `full`, `thorough`, and `extreme`. Milestone-level
robustness takes precedence over driver-level robustness when both are
present; `north_star_critical` is OR'd across the chain-level and
milestone-level settings.
### Missing packaged skill paths
Some North Star action types conceptually suggest packaged skills (e.g. a
`dead_delete` action could map to a dead-code-removal skill), but no such
skills ship with the pipeline today. These are **packaging follow-up
items** — the pipeline enforces the structured action contract regardless,
and any future skill packaging would consume the existing
`north_star_actions[]` / `north_star_actions_addressed[]` metadata without a
contract change.
View on GitHub