complexity-first-planning

Part of the Reasoning Discipline Protocol. Engage per rdp-gate (skip on trivial / linear tasks; light touch on a strong-reasoning host).

Provenance. This is an RDP derivation from general engineering discipline (risk-first / critical-path / pre-mortem) — it is not an Anthropic- documented Fable behavior. Fable's "start at the top of your difficulty range" is about task selection (give the model harder tasks), not intra-task order. The skill stands on its own merit; it is not sold as a frontier-model transplant.

When to use

• Staging multi-component work where the hardest/most-uncertain part is not yet proven.
• A plan whose later steps depend on an assumption that could collapse.

Do NOT use for single-step, linear, or fully-specified tasks (no load-bearing unknown to resolve), or when the user has already fixed the sequence.

When the agent should load this

• The user asks to "plan", "break down", or "stage" work that spans ≥2 components and at least one part is unproven.
• A multi-step plan is forming whose later steps assume something untested.
• Mid-task: a step just failed because an earlier, easier step baked in a wrong assumption — reload this and re-sequence risk-first.

Procedure

1. Inspect and name the unknowns. Read the affected components first (start from the repo slot of the context-spine when present), then list which carry real uncertainty (technical feasibility, an unverified integration, an ambiguous requirement) — analyze the existing system before planning any change.
2. Assess and rank by load-bearing risk. The load-bearing unknown is the one whose failure invalidates the most dependent work — not the one that is merely hard.
3. Resolve it first. Spike / probe / prototype the load-bearing unknown before building anything that depends on it. Record the result in the notes file (see notes-first-reasoning): prediction → result → lesson.
4. Cascade. Once the riskiest assumption holds (or is corrected), sequence the dependent work. If it fails, the cheap early failure saved the rework.

Output

A short ordered plan that leads with the load-bearing unknown + how it will be proven, then the dependent steps. One recommendation, not a survey.

The four plan criteria — what a reviewer fails the ordered plan on

Risk-first ordering is one of four things a plan is judged on. All four are stated so a reviewer can fail a plan on them — each names what failing looks like. This skill owns the third; the other three are stated here because a plan reviewed on ordering alone passes while being unnecessary, insufficient, or built on verbs that do not exist. The same four are carried in feature-planning (pack product-basic), which is pruned under projection.mode: scoped and so is named rather than linked.

1. Necessity — every step traces to a stated problem or acceptance criterion. Fails when the plan carries work nothing asked for — a spike for an unknown that is not load-bearing is the version of this failure this skill causes.
2. Sufficiency — every requirement is discharged by at least one step. Fails when a requirement owns no step, or its only step has no verifiable outcome. Walk the requirements, not the steps.
3. Ordering — the load-bearing unknown is resolved before anything that assumes it, and no step consumes what a later step creates. Fails when the plan front-loads the easy parts, or ranks by effort instead of dependency blast radius (see Gotchas). This is the criterion this skill exists for.
4. Groundedness — every command, skill and tool the plan tells someone to run exists. Fails when the plan names a verb that is not in the index. Check it against CAPABILITIES.yaml at the repository root — the generated capability index of every shipped skill and command — plus the project's own script/task entry points. It is the only one of the four decidable without judgement, so it is never waved through.

Do NOT

• Build the easy parts first to show progress, then discover the hard part breaks them.
• Treat "hardest" as "most code" — rank by dependency blast radius, not effort.
• Over-plan a strong-reasoning host (it sequences risk natively — keep it light).

Gotchas

• Mistaking effort for risk. A 400-line but well-understood refactor is low load-bearing risk; a 5-line call into an unverified third-party API is high. Ranking by size instead of dependency blast radius is the classic failure.
• "Resolved on paper". Reasoning that the unknown "should work" is not resolving it — the spike must actually run / compile / return before dependent work starts. Record prediction → result, not prediction → assumption.
• Spike sprawl. The probe answers exactly one question (does the load-bearing assumption hold?), then stops. Turning it into the real implementation defeats the cheap-early-failure purpose.

Related Skills

WHEN to use this

• Staging multi-component work where the hardest / most-uncertain part is unproven.
• A plan whose later steps rest on an assumption that could collapse.

WHEN NOT to use this

• Single-step, linear, or fully-specified work — no load-bearing unknown to resolve; the rdp-gate filters these.
• Breaking a feature into tasks in general — route to feature-planning, which composes this skill.
• Recording the spike's prediction / result / lesson — that belongs in notes-first-reasoning.