Constructdocs
section · guides

Template Prompt Audit

3 min read·guides / concepts / template-prompt-audit

Template & prompt quality audit

Every document template (templates/docs/*.md) and specialist prompt (specialists/prompts/cx-*.md) judged against doc-quality-rubric.md: strong (an expert would respect it), adequate (sound structure, missing a depth dimension), thin (placeholder-level, not shippable). The gap column names what a domain expert would add; the standard column anchors it.

This is the assessment that drives remediation — re-verify a row when you touch it rather than trusting the label.

Templates

TemplateGradeKey gap vs expert standardStatus
prd, prd-platform, prd-business, meta-prdstrong
adr, rfc, rfc-platformstrongstructure now enforced
test-plan, memo, onboardingstrong
research-briefstrongAdmiralty grade added; structure enforced
incident-reportthin → strongtrigger/root-cause/contributing-factors split, severity rationale, mitigators, action priorities, glossary (Google SRE)done (7zrh.5)
skill-artifactthin → strongcompetency rubric, prerequisites, failure modes, worked exampledone (7zrh.5)
research-findingthin → strongAdmiralty-graded sources, observation/inference split, confidence reasoning, refreshdone (7zrh.5)
runbookadequateSLO/severity-to-action mapping, diagnostic decision tree (Google SRE error budgets)queued (Phase C)
strategyadequateresources, leading/lagging metrics, milestones, risk register, kill criteriaqueued (Phase C)
evidence-brief, signal-brief, product-intelligence-reportadequateconfidence calibration + evidence bars; sources table (evidence-brief done)queued (Phase C)
prfaq, customer-profile, one-pager, changelog-entry, persona-artifact, backlog-proposaladequateevidence grounding / decision-forcing depth per familyqueued (Phase C)

Specialist prompts

Depth is split between the prompt and the skills/roles/*.md overlay; the gap is usually a named methodology framework missing from the overlay, not a shallow prompt.

SpecialistGradeMethodology gapStatus
cx-architect, cx-engineer, cx-ai-engineer, cx-reviewer, cx-qastrong
cx-securitystrongmake STRIDE/PASTA process explicitqueued (Phase D)
cx-srestrongmake error-budget policy explicit (Google SRE)queued (Phase D)
cx-researcher, cx-product-manager, cx-data-analyst, cx-docs-keeperstrongsource taxonomy added to cx-researcher
cx-evaluatoradequate → strongrubric design, ground-truth/inter-rater reliability, FP/FN cost asymmetry, statistical significancedone (7zrh.6)
cx-orchestratoradequate → strongdependency-graph wave sequencing, critical path, fan-out boundingdone (7zrh.6)
cx-ux-researcheradequatesampling/power, validity types (internal/external/construct), inter-rater reliabilityqueued (Phase D)
cx-data-engineeradequatedata lineage/observability, SLA maturityqueued (Phase D)
cx-business-strategistadequatePorter's Five Forces, scenario planningqueued (Phase D)
cx-legal-complianceadequaterisk taxonomy, liability framing, technical-legal bridgequeued (Phase D)
cx-debuggeradequatecausal-model root-cause enumerationqueued (Phase D)
cx-devil-advocateadequateFMEA / failure-mode enumerationqueued (Phase D)
cx-operationsadequatecritical-path method, resource levelingqueued (Phase D)
cx-rd-leadadequatepower analysis, effect-size estimationqueued (Phase D)
cx-release-manageradequatecanary failure-detection, SLO-based rollback treesqueued (Phase D)
cx-trace-revieweradequatestatistical process control, drift detectionqueued (Phase D)
cx-platform-engineeradequateIaC maturity model, supply-chain SBOMqueued (Phase D)
cx-designer, cx-accessibilityadequatedesign-system maturity; screen-reader/cognitive-accessibility rigorqueued (Phase D)
cx-exploreradequatedependency-graph analysis, code-smell enumerationqueued (Phase D)

Sequencing

  • PR-1 (this branch) — foundation (source taxonomy, quality rubric, structure enforcement) + all thin items remediated.
  • PR-2 (Phase C) — adequate templates.
  • PR-3 (Phase D) — adequate prompts, mostly via role-overlay methodology additions.

References