| name | form-guidance |
| description | Decide the smallest truthful form for a repeated problem before designing its implementation: no new artifact, local instruction, durable decision record, skill, runtime, projection, or bounded campaign. Use when asking "should this be a skill?", "what form should this take?", "how should this capability exist?", "is this another agent/runtime?", "should we add a command or document?", or when a generic workflow would hide an authority, inheritance, or recurrence decision. Do not use for ordinary implementation after the form is already accepted. |
| argument-hint | [audit | decide] |
Form Guidance
Principle expression
Primary: P16
Supporting: P04, P14, P15
Scope
Own one judgment: what is the smallest form that lets the real actor make the
required recurring decision without creating false authority, duplicate memory,
or unnecessary machinery?
Do not begin from a preferred noun such as skill, agent, runtime, dashboard, or
plan. A form decision answers how a named domain judgment may exist, never
what that domain judgment should recommend. Do not implement the chosen form,
organize its paths, change the Sequence, define a domain methodology, or accept
its result. Route those responsibilities after the form decision and stop.
Principle source
Use a host Sequence and its matching interpretations when the host declares
them. Otherwise use this skill's read-only fallback in references/sequence.md.
Read only P16, P04, P14, and P15 for this
skill's stable lineage. A host task may select a different current lead, but it
does not create co-primary doctrine.
Start
State the form question before proposing an artifact:
Actor and next decision:
Observed recurring action gap:
Concrete evidence:
What must survive this session:
Who may verify and commit:
Strongest keep-as-is case:
If the action gap is one-off and no later actor must recover it, return a
conversation result. Do not manufacture a durable form.
Core method
- Ground the concrete need. Read the accepted design, actual runtime, and
evidence of repeated action or decision failure. Separate observed facts from
the proposed explanation. Name the actor whose next judgment is blocked.
- Name the principal form contradiction. Use P04 to identify the mismatch
whose resolution changes the available actions: for example, a durable
decision hidden in chat, a runtime capability exposed only as internal JSON,
or a recurring judgment re-invented every session. Keep secondary constraints
explicit.
- Compare the keep-as-is case with candidate forms. Use
the form matrix. A candidate must own a distinct
decision, lifetime, authority boundary, or runtime capability. It may be a
stack: a skill can prepare a decision artifact; a runtime can retain evidence;
a CLI can project either. Do not collapse those roles into one “central” form.
- Test authority and reconstruction. For each candidate, name its source,
verifier, committer, and whether it can be rebuilt. A view, summary, index,
dashboard, or generated plan is a projection unless explicitly promoted to a
governed source. P14 forbids giving it fact authority merely because it is
convenient.
- Choose the minimum valid form and owner. Use P15: retain only distinctions needed
to change the blocked decision while preserving hard constraints. Prefer an
existing owner when it already supplies the gate. Route to
skill-engineering for an agent-facing repeated judgment,
artifact-organization for authority-and-lifetime layout,
a project design decision for a new runtime capability, and
principle-cultivation for a possible new Sequence meaning. If the selected
form needs a new domain method, name that owner and its unsolved domain
question; do not draft its doctrine, plan, or workflow here.
- Make the decision actionable. Return a compact form decision: selected
form stack, non-selected alternatives, content owner, ownership map,
smallest next transition, acceptance observation, and what would disconfirm
it. Use the
record template only when handoff,
approval, or later recovery needs it.
- Verify the form, not its prose. Run an action probe, a boundary probe,
and a context/authority probe from evaluation.
A new form is only supported when the intended actor can use it, it declines
adjacent non-scope, and no projection silently becomes a source.
Form boundary
- A skill teaches an agent a repeatable judgment at activation time. It does
not own project facts, long-lived commitments, execution, or human approval.
- A durable decision artifact preserves an approved or reviewable judgment
across sessions. It does not itself execute work.
- A runtime owns real execution boundaries, state transitions, evidence, or
mechanical verification. It is justified only when a document or skill cannot
provide that capability.
- A projection improves access to a declared source and must be rebuildable.
- A bounded campaign record coordinates one multi-session transition, then
is promoted, archived, or discarded. It is not a permanent institution.
- A new domain method remains unformed after this Skill selects its carrier.
Its problem definition, concepts, and decision method belong to its future
owner; form-guidance records the handoff and does not fill that role.
Completion standard
The form decision is ready when it names the action gap, strongest unchanged
alternative, selected form stack, authority/source map, smallest next
transition, and a falsifiable probe. If recurring need, ownership, or durable
decision consequence is unproven, retain the current form and record the
uncertainty rather than creating a new skill or system.