dk-flavored-spec-kit - SKILL.md Agent Skill

name: dk-flavored-spec-kit description: "Shared knowledge base for the dk.speckit.* agent family. Defines evidence-first execution, risk classification, pushback protocol, verification strategy, and artifact contracts. Agents reference this skill to stay DRY — philosophy and formats live here, agent files focus on workflow." risk: safe author: danieldekay date_added: "2026-03-12"

dk-Flavored Spec-Kit

Shared knowledge base for the dk.speckit.* agent pipeline. This skill extracts all cross-cutting concerns so individual agent files stay focused on their workflow.

Philosophy

The dk-flavored spec-kit extends the OOTB spec-kit pipeline with three principles borrowed from Anvil:

Evidence-first execution — verification is recorded, not self-reported. If it's not in the ledger, it didn't happen.
Risk-scaled depth — not every task deserves the same scrutiny. Green tasks get a quick check; red tasks get adversarial review.
Pushback as a feature — agents that can say "this is a bad idea" before implementing save more time than agents that implement the wrong thing fast.

Agent Family

Agent	Role	Key Anvil Enhancement
`dk.speckit.main`	Lifecycle orchestrator (GPT-5)	Git hygiene, pushback triage
`dk.speckit.orchestrator`	State detection & routing (Sonnet)	Pushback routing, ledger-aware gates
`dk.speckit.tasks`	Task breakdown with risk labels	🟢🟡🔴 risk + S/M/L sizing
`dk.speckit.implement`	Anvil-enhanced execution	Verification ledger, pushback, reuse detection
`dk.speckit.quality-gate`	Automated checks	Unchanged — reads code
`dk.speckit.autofix`	Bounded fix loops	Ledger-aware context
`dk.speckit.review`	Evidence-backed review	Consumes evidence bundle, adversarial pass for 🔴
`dk.speckit.commit`	Conventional commits	Unchanged
`dk.speckit.retro`	Compound learning	Ledger analytics
`dk.speckit.ui`	UI requirements analysis	Unchanged

How Agents Reference This Skill

Agents load specific reference files before execution:

## Skill References

Before execution, read the following from `skills/dk-flavored-spec-kit/references/`:
- `risk-classification.md` — for interpreting 🟢🟡🔴 labels and S/M/L sizing
- `evidence-ledger.md` — for ledger format and logging rules
- `pushback-protocol.md` — for when/how to escalate via askQuestions

This keeps agent files focused on WHAT to do. The skill defines HOW and WHY.

Pipeline Flow

User Request
    │
    ▼
dk.speckit.main / dk.speckit.orchestrator
    │ (git hygiene → state detection → routing)
    ▼
SPECIFICATION STAGES (OOTB sk-v1, unchanged)
    constitution → specify → clarify → plan → tasks
    │                                         │
    │                          dk.speckit.tasks wraps
    │                          with risk + sizing
    ▼
dk.speckit.implement
    │ (ledger, risk-scaled verify, pushback ↔ orchestrator)
    ▼
dk.speckit.quality-gate ↔ dk.speckit.autofix
    │
    ▼
dk.speckit.review (evidence audit + adversarial for 🔴)
    │
    ▼
dk.speckit.commit → dk.speckit.retro (ledger analytics)

Artifact Flow

tasks.md (with 🟢🟡🔴 labels)
    │
    ├──→ dk.speckit.implement reads risk labels
    │         │
    │         ├──→ evidence-ledger.json (written per task)
    │         └──→ pushback-log.json (if escalations occurred)
    │
    ├──→ dk.speckit.review reads evidence-ledger.json
    │         │
    │         └──→ review-report.md (includes Evidence Audit pass)
    │
    └──→ dk.speckit.retro reads evidence-ledger.json
              │
              └──→ retro.md (includes ledger analytics)

Reference Index

Reference	Purpose	Used By
risk-classification.md	🟢🟡🔴 definitions, S/M/L sizing	tasks, implement, review
evidence-ledger.md	JSON ledger schema, logging rules	implement, review, retro
pushback-protocol.md	When/how to pushback, askQuestions flow	implement, main, orchestrator
verification-strategy.md	Depth scaling by risk level	implement, review
artifact-registry.md	All artifacts, ownership, locations	all agents