project-groundwork

Project Groundwork

A draft document is not settled just because it exists; internal contradictions and unresolved TBDs hide in it until implementation runs into them. This skill's defining constraint, from ISO/IEC/IEEE 29148's requirements quality characteristics: a specification is checked against a fixed, named set of properties rather than a vague sense of whether it "looks done." The standard distinguishes properties of one requirement (unambiguous, verifiable) from properties of the document as a whole (complete, consistent); both are checked here.

This is a narrow, existing-document job, distinct from three doctrine skills it would otherwise overlap:

• wayfinder charts open decisions when no document exists yet, or the whole effort is too large to spec in one pass. This skill starts from a draft that already exists.
• grilling is the general-purpose interview technique for closing any known-versus-needed gap. This skill uses grilling's batching discipline for the specific gaps this skill's own scan surfaces; it does not redefine the interview mechanics.
• domain-modeling owns the glossary and ADR format that decisions get recorded into over time. This skill's distinct output is writing the resolved decision into the source document itself, in the same step, not only into an ADR.

Phase 1: Scan against the quality characteristics

Read the draft and check it against each of these, noting concrete violations, not general unease:

• Unambiguous: does any section allow more than one reasonable interpretation?
• Consistent: do any two sections state something that cannot both be true?
• Complete: is there a TBD, a placeholder, or a decision the document assumes was made elsewhere but never states?
• Verifiable: is there a claim or requirement with no way to confirm it was met?

List each violation with its location, not a summary judgment that the document "needs work."

Phase 2: Separate blocking gaps from ones that can wait

Not every violation found in Phase 1 blocks implementation from starting. A genuinely blocking gap is one where guessing wrong would cost real rework; a non-blocking one can be left as a stated open question and revisited later. Sort the list before asking anything, since a batch of ten questions when only three are load-bearing wastes the reader's attention on the other seven.