Agent
Convenes the Fable 5 + Sol expert-consensus seat the product law names as the pipeline-wide default at any point that would otherwise ask the operator. Spawn when the blocker ladder reaches it (two provably-distinct fix attempts have both failed) or at any other agreement-pass trigger the product law or the…
Agent
Turns a sealed brainstorm transcript into the operator's project vision.md and decisions.jsonl — the brainstorm phase's exit artifacts (phase-2 exitemits, machine/phases-registry.json). This is the context-wall enforcer between the sketchbook's long dialogue and the rest of the pipeline. Spawn once, in one of three…
Agent
Runs the operator's brainstorm phase (phase 2, machine/phases-registry.json) — spawned by Kiln as an agent-team TEAMMATE (its own context, its own mailbox, the operator converses in its window) when agent teams are enabled; when they are not, this seat runs nowhere at all — Kiln facilitates the sketchbook itself…
Agent
Reviews every chunk implementer-opus.md produces — the routing config's chunkreviewofclaudework seat (GPT-5.6-Sol, effort medium, scoped) — a Sonnet 5 dispatcher hosting Sol at medium effort via Codex CLI, the same one-agent-hosts-two-seats pattern agents/plan-coverage-checker.md uses. Enforces the cross-family…
Agent
Reviews every chunk implementer-sol.md produces — the routing config's chunkreviewofgptwork seat (Sonnet 5), enforcing the cross-family chunk-review law (the reviewer's family must differ from the implementer's). Diff-scoped correctness against git diff HEAD plus any untracked new file (never a whole-repo re-review)…
Agent
Kiln's delivery-voice renderer (the product law: "delivery: styled final presentation + the last-mile report — TWO artifacts"; the voice law, product/voice/kiln.md — fact first, one breath of style; no recorded event, no line; red is red). Spawned once, at the delivery phase, only after machine/src/delivery.js's…
Agent
The post-green exploration seat the verification architecture names last — "Discovery agent: roams only after green, time-boxed, findings become journeys or fix tasks — never gates." Spawned only once verification/suite-state.json reports every criterion green (or quarantined-and-accepted), never before; receives that…
Agent
The stop-writing seat for the handoff pair (handoff.md plus state.json's own stop fields) — a purpose-named seat backed by one handoff.js write-stop machine command, holding Bash for that command alone; Kiln supplies the operator's exact question or rationale as structured data and does not write or paraphrase it.…
Agent
The UI/creative chunk-implementation seat (routing config's implementeruicreative seat — Opus 5) for milestones architecture/milestones.json marks UI/creative in designerfamily. Same contract as implementer-sol.md, run natively rather than dispatched — no Codex hop, since Opus is the model doing the work directly…
Agent
The default chunk-implementation seat (routing config's implementer seat — GPT-5.6-Sol) for every milestone not marked UI/creative — Sonnet 5 dispatcher at high effort hosting Sol implementation sessions via Codex CLI, workspace-write, the same one-agent-hosts-two-seats pattern agents/agreement-pass.md and…
Agent
The kept-warm seat for one milestone — owns milestones/ /spec.md's chunk decomposition (per chunk WHAT to build + WHAT to test, never HOW), stays alive through the milestone's whole lifecycle (resume-dont-respawn, spawned fresh only at milestone start), and answers implementer/reviewer consultations through the…
Agent
The cross-family CUMULATIVE-milestone review seat (the product law's "milestone review closes each slice," distinct from per-chunk review) — the routing config's milestonereview: "cross-family" seat. Reviews a milestone's whole cumulative diff (git diff ..HEAD) plus its accumulated JIT journey results for slice-level…
Agent
The scripted-mechanics seat for onboarding (phase 1, machine/phases-registry.json's onboarding entry) — the seat that owns machine/src/preflight.js, which has no other caller in the roster. Spawned once, at onboarding entry, with a brief naming directly and stating whether Kiln's own detect() call found .kiln/ absent…
Agent
Hosts the Sol-authored candidate for the Maximal confrontation protocol only — Sonnet 5 at medium effort dispatching GPT-5.6-Sol via Codex CLI on the same one-agent-hosts-two-seats pattern agents/agreement-pass.md and agents/plan-coverage-checker.md use. Produces the identical five-artifact set plan-author produces…
Agent
The single-author planning seat for the trio of planning protocols (routing config's plandraft seat — fable-5, fallback opus-5). Drafts THE plan alone from the crystallized vision pair and the research summary, never from raw streams; emits the five architecture content artifacts (plan.md, criteria-register.json…
Agent
The fresh-Claude adversarial seat for the Deep scrutiny and Maximal confrontation protocols — Opus 5 at high effort, spawned with no memory of how the plan was authored. Narrow mission is PURE breaking only — hidden coupling and false sequencing in the plan's own internal logic — never infeasibility or resource…
Agent
Hosts BOTH of the product law's Sol-authored planning missions (routing config's coveragecheck seat — GPT-5.6-Sol, effort high, scoped), dispatched via Codex CLI on the same one-agent-hosts-two-seats pattern agents/agreement-pass.md uses. Mission 1 (every protocol's closing step): lost-nothing coverage — compare the…
Agent
The maximal-confrontation selection seat (routing config's plandraft seat again — fable-5, fallback opus-5 — spawned fresh, never the same transcript as either candidate's author). Receives two ANONYMIZED bundle directories (plan-A/plan-B, authorship mechanically stripped by machine/src/anonymize.js) plus their…
Agent
Owns the research phase (phase 3, machine/phases-registry.json) — decomposes the crystallized vision+decisions pair into worker briefs, spawns research-worker at settings.json's researchwidth (2/3/5), steers by resuming rather than respawning, and closes the phase with research/index.json plus research/summary.md.…
Agent
Context-blind executor of exactly one research brief issued by research-chief — the Sonnet 5 seat spawned in parallel at researchwidth (2/3/5), each instance on a distinct subtopic. Verifies its brief validates against machine/src/validators.js's 8-field contract before researching anything; researches official…
Agent
The brownfield explorer the product law names first in onboarding — "Brownfield: scout-and-inventory pass, then create all artifacts, then re-enter where state says." Spawned once, at onboarding, only after Kiln's own detect() call has already classified the target as brownfield (this seat trusts that classification…
Agent
The light security lane the product law names alongside the milestone-review security paragraph — "Security lane, light: scripted scan step + security paragraph in milestone reviews + a brief operator conversation on depth; heavy security = custom-pipeline territory." This seat runs ONLY the scripted scan step, once…