io.github.cdeust/prd-spec-generator

Catch a hallucinated spec before it becomes code.
It checks that every symbol exists, every acceptance criterion traces to a requirement, and no performance claim passes without a measurement — deterministically, with no model in the checking loop.
Works in Codex, Gemini CLI, Claude Code, and CI. Runs on your machine.

What An Agent Asks · Getting Started · Pipeline · Tools · Verification · Calibration · Architecture · Zetetic Standard

_{One of three MCP servers that each run standalone — memory, code graph, spec verification. How they compose ↓}

Every AI agent that drafts a PRD eventually invents a function that doesn't exist, claims latency it can't measure, or writes acceptance criteria that don't tie back to the requirements they're supposed to test. The output sounds confident. It is not actionable. The next stage in the pipeline — code generation, ticket import, sprint planning — silently inherits the hallucination, ships it, and pays for it later.

AI Architect MCP Spec is a TypeScript MCP server that fixes this at the structural level. Its portable verifier gives Codex, Gemini CLI, Claude Code, CI, and other stdio MCP hosts the same deterministic Hard Output Rules and cross-section traceability checks. The full pipeline is a stateless reducer (step(state, result?) → next_state, action) driven by a host that can execute its orchestration actions; Claude Code is the only packaged full-pipeline host today. Sections are produced one at a time, validated before the host ever sees them, and every load-bearing claim is judged by a panel of genius reasoning agents drawn from zetetic-team-subagents against the codebase graph from ai-architect-mcp-codebase. The loop is closed: per-judge reliability is calibrated from history, retry budgets are derived from survival statistics, KPI gates are tuned against frozen baselines, and held-out partitions are mechanically sealed so no calibration result can be peeked at before evaluation.

What you get

• A spec that survives contact with code. Symbols are cross-checked against the real codebase graph, so a PRD cannot reference a function that was renamed last week.
• Traceability that is actually checked. Every acceptance criterion must tie back to a requirement; numbering gaps are caught, not assumed away.
• No PASS without a measurement. The verdict taxonomy refuses to pass a latency, throughput, fps or storage claim — it returns SPEC-COMPLETE or NEEDS-RUNTIME instead of a confident guess.
• A guard against agreeable judges. A panel that agrees 100% of the time over 5+ claims raises distribution_suspicious rather than being read as consensus.
• It works where you work. The deterministic verifier travels to Codex, Gemini CLI, Claude Code and CI; the full generation pipeline runs on hosts that can execute its orchestration actions.

Sovereign verification, eco-responsible by intent

Sovereign is what it is today. The server is a stateless reducer — step(state, result?) → next_state, action. It emits actions; your host executes them. It runs on your machine, and it never phones home for a verdict.

Eco-responsible is what we're aiming at. This one's lever is different from its siblings': it is avoided rework. A hallucinated PRD is inherited by code generation, ticket import and sprint planning before anyone notices — and every stage that consumed it has to be redone. The spec stage is the cheapest place in the pipeline to catch that, and three design choices keep the catching itself cheap:

• The checks run no model. Hard Output Rules and cross-section traceability are deterministic; finding a broken FR-AC link costs zero inference.
• Sections fail before they are finished. Each is validated before the host ever sees it, instead of being generated in full and then discarded.
• Retry budgets are derived, not guessed. MAX_ATTEMPTS comes from Kaplan-Meier survival math, and map_failure_to_retrieval sends a corrective query before a retry — a retry that reuses the context that just failed is waste with a receipt.

As in the sibling repos, we publish no energy or CO₂ figure: this repository measures no joules, and avoided rework is a design argument rather than a measurement.

10 packages. 17 MCP tools. 20 pipeline steps (11 PRD generation + 9 opt-in implementation). Multi-judge verification with consensus. Closed-loop calibration with externally-grounded falsifiers. 1506 tests. Every numeric constant traces to a citation, a benchmark, or a // source: provisional heuristic admission.

Closed-loop reliability calibration

Verification is not a one-shot pass/fail report. Every claim resolution can flush an observation back to a calibration repository, every consensus run can pull calibrated posteriors from history, and every closed loop runs an external control arm so the calibration's effect is measured, not assumed.

• Per-judge Bayesian reliability calibration — Beta(7,3) prior with sensitivity / specificity split per claim_type. Posteriors stored in a SQLite-backed ReliabilityRepository; observations flushed on every claim resolution.
• MAX_ATTEMPTS retry calibration — Kaplan-Meier survival math (kmEstimate / kmMedianAttempts / logRankTest with Greenwood + Brookmeyer-Crowley CIs); Schoenfeld sample-size derivation event-rate-corrected to ~519 (was 823) against the measured event_rate=0.4762, CP CI [0.4456, 0.5069].
• KPI gate tuning — Clopper-Pearson exact CIs; per-machine-class wall_time normalization with 5-bucket detectMachineClass; frozen-baseline content-hash assertion; loadCalibratedGates + hold_provisional ratchet protection.
• Plan-mismatch fire-rate — measured via XmR control charts (Wheeler 1995, Western Electric 1956) with a synthetic injection round-trip pre-flight that catches drift between the diagnostic prefix and the regex matcher.
• Externally-grounded held-out subsets — Ajv schema oracle, mathjs oracle, tsc subprocess code oracle, validateSection spec oracle. OracleUnavailableError typed throw replaces stub-mode fabrication. This is the layer that breaks annotator-circularity — judges and oracles share no inference path.
• CC-3 forced-exploration control arms — every closed loop carves out a 20% partition that reverts to the prior. Without it, calibration-on-calibration looks like progress whether or not it actually is.
• Cross-arm comparison metrics — computeAblationComparison / computeReliabilityComparison / computeKpiGateComparison produce paired-bootstrap CIs (Efron & Tibshirani 1993 §16.4; deterministic mulberry32 RNG; 12-decimal reproducibility pin). Outcome is a falsifiable recommendation: calibrated_helps, prior_helps, or inconclusive_underpowered.
• Mechanically-enforced held-out partition seals — three sealed lock files (maxattempts-heldout.lock.json, kpigates-heldout.lock.json, heldout-partition.lock.json) commit a sha256 of the partition before evaluation. The SEAL_VERIFIED typeof sentinel is the only way to compute cross-arm metrics on a sealed partition; passing anything else is a type error at the boundary.
• Production-mode dispatcher — makeProductionDispatcher + AgentInvoker interface. The CLI --mode production|canned flag selects whether calibration sees real verdicts or canned ones; the canned arm is preserved for offline reproducibility.

What an agent can ask it

start_pipeline(feature_description, codebase_path?)
  → returns the first NextAction; the host executes it and feeds the result
    back via submit_action_result. Eleven steps later: 9 PRD files written.

submit_action_result(run_id, result)
  → drives the reducer one more step. The host sees only SUBSTANTIVE actions
    (ask_user, call_pipeline_tool, call_cortex_tool, spawn_subagents,
     write_file, done, failed). emit_message is coalesced into the
     messages array; the host never has to "advance past" a banner.

validate_prd_section(content, section_type)
  → deterministic Hard Output Rules — zero LLM calls, pure regex/parsing.
  → returns: violations[], hasCriticalViolations, totalScore.

validate_prd_document(sections[])
  → cross-section checks: SP arithmetic, AC numbering, FR-AC coverage,
    test traceability. Catches what per-section validation misses.

coordinate_context_budget(prd_context, completed_sections[])
  → per-section retrieval/generation token budgets so Cortex recall and
    section drafting don't fight over the same context window.

map_failure_to_retrieval(violations[])
  → closes the validation→retrieval feedback loop. When a section fails
    validation, this returns the corrective Cortex query that would
    have prevented the failure.

Getting started

_{A note on naming. This is published as ai-architect-mcp-spec across its Claude Code, Codex, Gemini, MCPB, PyPI/npm and MCP Registry surfaces. 0.7.0 moved the Registry entry and primary bundle to that identity; 0.8.0 completed the migration, so no prd-spec-generator plugin or release alias is retained anywhere. The former io.github.cdeust/prd-spec-generator Registry versions are deprecated with a pointer to the canonical entry rather than left active and frozen.}

Install (marketplace — recommended)

claude plugin marketplace add cdeust/ai-architect-mcp-spec
claude plugin install ai-architect-mcp-spec@ai-architect-mcp-spec-marketplace

Restart your Claude Code session. The 17 MCP tools register on first stdio handshake. Then:

/generate-prd build OAuth login for the admin console

The plugin's bundled MCP server at mcp-server/index.js is self-contained (only better-sqlite3 is an optional native dependency for the evidence repository — gracefully degrades to in-memory mode when absent).

Migrating from prd-spec-generator

The old plugin and marketplace are deprecated publications, not aliases. Remove them before installing the canonical package:

claude plugin uninstall prd-spec-generator@prd-spec-generator-marketplace
claude plugin marketplace remove prd-spec-generator-marketplace
claude plugin marketplace add cdeust/ai-architect-mcp-spec
claude plugin install ai-architect-mcp-spec@ai-architect-mcp-spec-marketplace

Callers that used prd-spec-generator:generate-prd must use ai-architect-mcp-spec:generate-prd. The prd-gen MCP server name, @prd-gen/* private workspace packages, PRD_GEN_* environment variables, and .prd-gen local data directory remain stable internal interfaces; they do not keep the deprecated publication installed.

Plugin-scoped MCP tool names include both the publication name and the stable server key, so prompts, automation, and allowlists must migrate the complete prefix:

• old: mcp__plugin_prd-spec-generator_prd-gen__<tool>
• new: mcp__plugin_ai-architect-mcp-spec_prd-gen__<tool>

Gemini CLI also keys installed extensions by the manifest name. Replace an installed extension under the former identity instead of updating it in place:

gemini extensions uninstall prd-spec-generator
gemini extensions install https://github.com/cdeust/ai-architect-mcp-spec

Companion ecosystem

For full effect, install the three companion plugins so the pipeline can consume codebase intelligence, persistent memory, and the genius-agent panel:

claude plugin marketplace add cdeust/ai-architect-mcp-codebase  # codebase graph intel
claude plugin marketplace add cdeust/Cortex                  # persistent memory
claude plugin marketplace add cdeust/zetetic-team-subagents  # the genius + team agents

claude plugin install ai-architect-mcp-codebase
claude plugin install hypermnesia-mcp
claude plugin install zetetic-team-subagents

Each plugin is independently useful; together they are the ai-architect ecosystem. See Companion ecosystem above.

Building from source

For development or to run the audit cycle locally:

git clone https://github.com/cdeust/ai-architect-mcp-spec.git
cd ai-architect-mcp-spec
pnpm install --frozen-lockfile
pnpm build      # builds all 9 buildable packages via tsc
pnpm bundle     # produces the standalone mcp-server/index.js
pnpm test       # 1506 tests (vitest workspace, all packages; the
                # ai-architect-mcp-codebase wire contract is pinned
                # unconditionally — see docs/INTEGRATION-TESTING.md)

pnpm verify runs all of the above (install + build + bundle + test) — same as CI.

Prerequisites for source builds: Node.js 20.x or 22.x, pnpm v10+ (corepack enable && corepack prepare pnpm@10).

Smoke-test offline

# Reducer end-to-end without a real host (uses the canned dispatcher):
pnpm test --filter @prd-gen/orchestration smoke

# Benchmark KPI run:
pnpm test --filter @prd-gen/benchmark pipeline-kpis

Both run in <2s on an M-series Mac. No LLM calls, no MCP traffic — the reducer is fully driven by canned ActionResults so you can audit behaviour offline.

Use with other MCP hosts

Honest scoping first. The full pipeline is host-dependent: the reducer returns actions (ask_user, spawn_subagents, write_file) that assume a host which executes them and feeds results back. Claude Code (via the plugin + /generate-prd skill) remains the only packaged host for that workflow.

Codex and Gemini instead receive the portable Spec Verifier: a no-LLM surface with exactly two deterministic tools and the same audit-prd and validate-spec skills:

• validate_prd_section — Hard Output Rules against a single section
• validate_prd_document — cross-section checks (SP/AC/FR/test traceability)

Install it in Codex from the repository marketplace:

codex plugin marketplace add cdeust/ai-architect-mcp-spec
codex plugin add ai-architect-mcp-spec@ai-architect-mcp-spec-marketplace

Install the same package as a Gemini CLI extension:

gemini extensions install https://github.com/cdeust/ai-architect-mcp-spec

Both manifests launch mcp-server/index.js with --profile verifier. The existing Claude manifest passes no profile and therefore retains the default 17-tool full surface.

A zero-violation result means that the document satisfies the implemented structural rules. It does not prove factual accuracy, product value, implementation feasibility, security, or semantic correctness.

For an un-packaged MCP host, clone the repository and select the same narrow profile explicitly. Replace /abs/path with the clone location.

git clone https://github.com/cdeust/ai-architect-mcp-spec.git /abs/path/ai-architect-mcp-spec
cd /abs/path/ai-architect-mcp-spec/mcp-server
npm ci --omit=dev

OpenAI Codex CLI (~/.codex/config.toml, shared with the ChatGPT desktop app and Codex IDE extension):

codex mcp add prd-spec -- node /abs/path/ai-architect-mcp-spec/mcp-server/index.js --profile verifier

Cursor (.cursor/mcp.json) and Windsurf (~/.codeium/windsurf/mcp_config.json):

{
  "mcpServers": {
    "prd-spec": {
      "command": "node",
      "args": [
        "/abs/path/ai-architect-mcp-spec/mcp-server/index.js",
        "--profile",
        "verifier"
      ]
    }
  }
}

VS Code (.vscode/mcp.json):

{
  "servers": {
    "prd-spec": {
      "type": "stdio",
      "command": "node",
      "args": [
        "/abs/path/ai-architect-mcp-spec/mcp-server/index.js",
        "--profile",
        "verifier"
      ]
    }
  }
}

The pipeline

The reducer produces twenty sequential steps: eleven PRD-generation steps, then nine opt-in post-specs implementation steps entered only through a human gate. Each step emits at most one substantive action; the host executes it and feeds the result back. A typical trial-tier feature run (11 sections) takes ~62 host-visible iterations.

Steps 12–20 are the opt-in post-specs implementation loop: implementation_gate (human gate — "Implement" vs "PRD only"; also writes 10-verification-report.md) → pre_impl_grounding → implementation (engineer subagent in an isolated worktree) → post_impl_verification (index → detect changes → semantic diff → security gates) → testing → review (a FAIL verdict retries implementation on the same worktree, bounded by REVIEW_RETRY_CAP) → pr_gate (mandatory human gate before any push) → pr_creation → finalize. Answering "PRD only" at implementation_gate skips straight to finalize; complete is the terminal marker, not a step.

Every step is independently testable (stepOnce(state, result?) returns the same shape as the runner). The runner coalesces emit_message actions internally so the host never sees a no-op.

The MCP tools

Three surfaces. The reducer drives the full pipeline; the validation + verification

• budget tools can be consumed directly by other systems without entering the pipeline; the diagnostics surface exposes config + health + history.

Reducer (3):
  start_pipeline             Initialize a run; returns first NextAction
  submit_action_result       Drive the reducer one step; returns next NextAction
  get_pipeline_state         Read-only state snapshot for diagnostics

Validation (2):
  validate_prd_section       Hard Output Rules — single section
  validate_prd_document      Cross-section checks (SP/AC/FR/test traceability)

Verification (3):
  plan_section_verification  Extract claims + select judge panels
  plan_document_verification Same, document-wide
  conclude_verification      Aggregate JudgeVerdict[] → VerificationReport;
                             accepts optional `claims` array carrying
                             `external_grounding` so oracle-resolved ground
                             truth can replace LLM-only consensus where
                             schema/math/code/spec oracles are available

Budget + feedback (2):
  coordinate_context_budget  Per-section token allocation
  map_failure_to_retrieval   Validation failure → corrective Cortex query

Diagnostics (7):
  get_config, read_skill_config, check_health, get_prd_context_info,
  list_available_strategies, get_quality_history, get_strategy_effectiveness

Each tool takes structured Zod-validated arguments and returns a typed response. No tool calls an LLM — section drafts and judge verdicts come back via the host's spawn_subagents action so the same pipeline runs against any agent runtime.

Multi-judge verification

The self_check step is a two-phase contract. Phase A plans the verification batch and persists a snapshot of (claim_ids, judges) to state. Phase B receives the verdicts, parses them against the snapshot, and aggregates via the consensus engine.

plan_document_verification(sections[])
  → extracts atomic Claims (FR-001, AC-005, NFR-LATENCY-1, ...)
  → selects a panel per claim type:
      architecture        → liskov + alexander + dijkstra + architect
      performance         → fermi + carnot + curie + erlang
      security            → wu + ibnalhaytham + security-auditor
      data_model          → mendeleev + dba + lavoisier
      acceptance_criteria → toulmin + popper + test-engineer
      ...

[host spawns the panel; each agent returns a JSON verdict]

conclude_verification(verdicts[])
  → Per claim, runs consensus():
      strategy: weighted_average (default) | bayesian
      fail_threshold: 0.5  (≥50% confidence-weighted FAIL → forces FAIL)
      precautionary tie-breaker: more-severe verdict wins
  → distribution_suspicious flag fires when 100% PASS over ≥5 claims
  → returns ConsensusVerdict[] with full distribution + dissenting list

The verdict taxonomy is deliberately five-level — not binary. NFR claims (latency, fps, throughput, storage) MUST NOT receive PASS: they are SPEC-COMPLETE if a measurement method is specified, NEEDS-RUNTIME otherwise. Judges that default to PASS for everything are caught by the distribution_suspicious detector and flagged in the typed done.verification field.

Calibration & falsification

The verification subsystem is itself a hypothesis: that consensus weighted by historically-calibrated reliability outperforms consensus weighted by a uniform prior. The closed loop below is what tests it.

1. Observe. Each verification run can flush per-judge observations (claim_id, claim_type, judge_id, verdict, oracle_truth?) to a SQLite reliability repository. Observations carry an external_grounding field that propagates from Claim through the orchestrator to the oracle resolution path; when an external oracle (Ajv schema, mathjs, tsc, validateSection) can resolve the claim, its truth replaces LLM-only consensus.
2. Calibrate. On subsequent runs, calibrated posteriors weight consensus per judge, per claim_type. A 20% control-arm partition is forced-explored using the prior (getReliabilityForRun / getRetryArmForRun decide which arm a given run lands in deterministically from run_id). Without the control arm, calibration-on-calibration looks like progress whether or not it actually is.
3. Compare. Cross-arm metrics (computeAblationComparison, computeReliabilityComparison, computeKpiGateComparison) run paired-bootstrap CIs (Efron & Tibshirani 1993 §16.4; deterministic mulberry32 RNG; 12-decimal reproducibility pin) and emit one of three falsifiable recommendations: calibrated_helps, prior_helps, or inconclusive_underpowered.
4. Seal. Held-out partitions are committed to lock files (maxattempts-heldout.lock.json for §4.2, kpigates-heldout.lock.json for §4.5, heldout-partition.lock.json for the §4.1 50-claim externally-grounded corpus) with a sha256 hash of the partition. The cross-arm metric functions accept a SEAL_VERIFIED typeof sentinel as a parameter; the only way to obtain that sentinel is to verify the seal first. Peeking at a held-out partition before evaluation is a type error.
5. Ground. Where an external oracle can resolve a claim deterministically, it does. Where it cannot, OracleUnavailableError is thrown rather than fabricating a stub-mode truth. This is the line that breaks annotator-circularity: judges trained against (or biased toward) LLM-style reasoning cannot poison calibration that uses non-LLM truth.

The lock files, the seal-verification dance, and the control-arm partition together mean: when a cross-arm comparison says "calibrated_helps with 95% CI excluding zero," the claim is measured, not vibes-checked. When it says "inconclusive_underpowered," that is also a falsifiable claim — you need more data, not more confidence.

Architecture

Ten workspace packages, each independently buildable, with strict Clean Architecture layering enforced by package boundaries.

core              ← domain types, schemas, agent identities
                    │  no I/O, no infrastructure dependency
                    │  Zod-validated; the only place where verdict /
                    │  section_type / capability shapes are defined
                    ▼
validation        ← Hard Output Rules (per-section + cross-section)
                    │  pure functions; no I/O
                    ▼
strategy          ← thinking-strategy selector (genius pattern routing)
                    │
meta-prompting    ← prompt builders for clarification / draft / jira
                    │  pure string composition
                    ▼
verification      ← claim extraction + judge selection +
                    │  consensus engine (weighted_average + Bayesian)
                    │  + buildJudgePrompt
                    ▼
orchestration     ← stateless reducer, 20 step handlers, runner
                    │  step(state, result?) → next_state, action
                    │  emit_message coalescing; canned-dispatcher utility
                    ▼
ecosystem-adapters← StdioMcpClient, AiArchitectCodebaseClient, CortexClient
                    │  the only package allowed to do I/O
                    ▼
mcp-server        ← composition root; 17 tools registered;
                    │  evidence repository (better-sqlite3, optional)
                    ▼
benchmark         ← pipeline KPI measurements + golden-fixture HOR scoring
                    │  + calibration/ subtree:
                    │    · ReliabilityRepository (SQLite, observation flush)
                    │    · Kaplan-Meier + log-rank + Schoenfeld N
                    │    · Clopper-Pearson exact CI + XmR control charts
                    │    · paired-bootstrap (Efron-Tibshirani 1993 §16.4)
                    │    · external oracles (Ajv / mathjs / tsc / validate)
                    │    · machine-class detector + frozen-baseline gates
                    │    · sealed held-out lock files (sha256 + SEAL_VERIFIED)
                    │    · production-mode dispatcher + AgentInvoker seam
                    │  Audit lineage: JSONL + .xmr sidecars per run.
skill             ← SKILL.md + slash-command definitions for Claude Code

Dependency rule (absolute)

Every package's package.json is checked: core depends only on zod; verification depends only on core; orchestration depends on core/validation/verification/meta-prompting (NOT on ecosystem-adapters); ecosystem-adapters depends on core + verification; mcp-server is the only place where everything composes.

A cross-audit found and fixed two layer violations:

• orchestration was importing extractJsonObject and buildJudgePrompt from ecosystem-adapters — pure utilities lived in the wrong package; moved to core and verification respectively.
• Pure domain types (Claim, JudgeVerdict, JudgeRequest, AgentIdentity) lived in ecosystem-adapters/contracts/subagent.ts; moved to core/domain/agent.ts. The infrastructure package now re-exports them as a backward-compat shim.

What this fixes that previous PRD generators don't

How it composes with the rest of the ecosystem

                            ┌────────────────────────────────────┐
                            │         Claude Code (host)         │
                            └────────────────────┬───────────────┘
                                                 │ stdio MCP
              ┌──────────────────────────────────┼──────────────────────────────────┐
              ▼                                  ▼                                  ▼
   ┌───────────────────┐              ┌────────────────────┐              ┌───────────────────┐
   │   ai-architect-   │   graph_path │   ai-architect-    │  recall      │      Cortex       │
   │   mcp-codebase    │ ───────────► │   mcp-spec         │ ◄─────────── │   (memory engine) │
   │   (Rust MCP)      │              │   (TS MCP)         │              │   (Python MCP)    │
   │                   │   symbols    │                    │  excerpts    │                   │
   │   read-only       │ ◄──────────► │   stateless        │ ───────────► │   thermodynamic   │
   │   intelligence    │              │   reducer          │              │   memory          │
   └───────────────────┘              └─────────┬──────────┘              └───────────────────┘
                                                │
                                                │ spawn_subagents
                                                ▼
                                  ┌─────────────────────────────┐
                                  │   zetetic-team-subagents    │
                                  │   97 genius + 19 team       │
                                  │   Each judge cites its      │
                                  │   primary paper.            │
                                  └─────────────────────────────┘

Each project owns one concern. ai-architect-mcp-codebase knows what's true about the code. Cortex knows what we already decided. zetetic-team-subagents knows how to reason about a specific shape of claim. AI Architect MCP Spec is the deterministic glue that turns those three signals into a PRD an agent can act on.

The Zetetic Standard

Every load-bearing constant in this codebase carries a // source: annotation. Three forms are accepted:

// source: <citation>          // a paper, a spec, a referenced design doc
// source: benchmark <path>    // a committed benchmark whose output produced this value
// source: provisional heuristic — <calibration plan>
                               // honest admission; tells the next reader
                               // (a) why the value is what it is today and
                               // (b) what evidence would change it

The cross-audit found and tagged every previously bare constant. Examples:

// pipeline-kpis.ts
const KPI_GATES = {
  /** source: provisional heuristic. Smoke baseline = 62 iterations on
   *  trial+codebase; cap is 100 (~60% headroom). dijkstra cross-audit
   *  derived a structural max of 9 emit_message hops; the substantive-
   *  action count builds on that. Replace with measured P95 + 1σ once
   *  the calibrated baseline has enough runs. */
  iteration_count_max: 100,
  ...
};

// verification/consensus.ts
/** source: provisional heuristic — Beta(7,3) (mean 0.7, ESS=10,
 *  moderately informative toward reliability). Replaced per agent by
 *  Beta(α+correct, β+incorrect) once calibrated from history. */
const DEFAULT_RELIABILITY_PRIOR_MEAN = 0.7;

The four pillars (consistent / true / useful / necessary) and the seven rules of zetetic inquiry are inherited from the zetetic-team-subagents standard. Provisional values are not silently propagated as truth.

What this system does not do

The same standard applied to itself.

1. The PRD pipeline does not write code — and the server itself never does. Steps 1–11 produce a PRD; symbols in the PRD are validated against the graph but never edited by us. The opt-in implementation loop (steps 12–20) runs only when a human answers "Implement" at implementation_gate; there, the host's engineer/test-engineer/code-reviewer subagents write, test, and review the implementation in an isolated worktree, and nothing is pushed or turned into a PR without the mandatory pr_gate approval. This server only emits spawn_subagents actions — it never edits source files or pushes branches itself.
2. It does not validate prose quality. Hard Output Rules check structural invariants (FR numbering, AC traceability, NFR shape, cross-references). They do not check whether a sentence is well-written or persuasive. That is what the multi-judge phase is for, and even there the judges return verdicts on claims — atomic assertions — not on style.
3. The judge phase is end-to-end testable but the judges are not deterministic. In tests we use a canned dispatcher that returns 100% PASS by construction; the distribution_suspicious detector exists precisely because real judge panels can also degenerate into confirmatory consensus, and we do not pretend otherwise.
4. The KPI gates are calibrated, not hand-picked. iteration_count_max, wall_time_ms_max, and mean_section_attempts_max are calibrated against the K=100 frozen baseline with Clopper-Pearson exact CIs, per-machine-class wall_time normalization, and loadCalibratedGates + hold_provisional ratchet protection. The lock file commits a content-hash of the baseline; mutating it post hoc fails the seal verification. Where data is still thin, gates remain hold_provisional rather than locked. See docs/PHASE_4_PLAN.md for the full pre-registration.
5. Citation presence ≠ citation validity. A // source: Knuth 1998 comment satisfies the convention whether or not Knuth 1998 exists or supports the value. We enforce that the citation IS THERE; the cross-audit cycle (genius + team review every phase) is what keeps it honest.

Reproducing the audit cycle

The repo ships a multi-agent cross-audit workflow. After every non-trivial phase:

# Engineering team review:
#   architect, code-reviewer, refactorer, test-engineer, security-auditor,
#   devops-engineer, dba (when relevant)

# Genius team review:
#   feynman (integrity), curie (measurement), popper (falsifiability),
#   dijkstra (correctness), shannon (signal), deming (variation),
#   poincare (qualitative), ...

Each agent reads the current state of the code (not from memory) and produces a ranked finding list, severity-ranked. Findings are closed in the cycle that raised them; anything that cannot be closed without new measurement is carried with the evidence required to close it written down, so a deferral is a stated obligation rather than a silence.

Project layout

packages/
├── core/                  Domain types · schemas · agent identities · evidence repo
├── validation/            Hard Output Rules · per-section + cross-section validators
├── verification/          Claim extraction · judge selection · consensus engine
├── meta-prompting/        Prompt builders (clarification / draft / jira)
├── strategy/              Thinking-strategy selector
├── orchestration/         Stateless reducer · 20 step handlers · runner · canned-dispatcher
├── ecosystem-adapters/    StdioMcpClient · AiArchitectCodebaseClient · CortexClient
├── mcp-server/            Composition root · 17 MCP tools registered
├── benchmark/             Pipeline KPI measurement · golden-fixture HOR scoring
│   └── calibration/       ReliabilityRepository · KM survival ·
│                          Clopper-Pearson · XmR · paired-bootstrap ·
│                          external oracles · sealed held-out partitions ·
│                          production-mode dispatcher
└── skill/                 SKILL.md · slash-command definitions

License

MIT — see LICENSE.

Project documents: ROADMAP (what the project intends to do, and what it will not do, for the next year) · CONTRIBUTING (how to propose a change, and the testing policy) · GOVERNANCE (who decides, and what happens to the project if the maintainer stops) · SECURITY (how to report a vulnerability, and the supply-chain guarantees) · Assurance case (the security argument and where it stops) · CODE OF CONDUCT · CHANGELOG.

This software is the independent work of Clément Deust. It was developed outside any employment relationship and is not affiliated with, endorsed by, or owned by any past or present employer. It is part of the ai-architect ecosystem (Cortex, ai-architect-mcp-codebase, zetetic-team-subagents).

Citations and primary sources for algorithms and constants are documented inline in the code via // source: annotations per the project's source-discipline rules. Removing those annotations does not change the license terms but does forfeit attribution to the original research the code is built on.

Don't ship a PRD that hallucinates a function it can't measure.
Ship one whose every claim was judged by Pearl, Curie, Liskov, and a panel of seven others, validated against the call graph, and grounded in what Cortex remembers from yesterday.