Proposal: Targeted Production-Hardening & Stochastic Guardrails for Core Technical Agents (Scoped by Division, 1 Commit per Agent)

[augustoheiss]

Thanks @msitarzewski for the thoughtful review on #1151 and #1237! Closing both to align on direction first makes complete sense — touching 86 agents (~6,500 lines) across the catalog is too wide a blast radius to review as code without agreeing on the boundaries first.

We have already rebased our work onto current main (b37b5bb, preserving all recent fixes from #1132–#1177, including engineering-devops-automator.md and the other 9 updated files) and split every single change into 1 atomic commit per agent (1 commit = 1 file), grouped by division so any individual agent can be evaluated, cherry-picked, or skipped independently alongside other contributors' fixes.

Here is the breakdown covering the four points you raised:

---

1. The Principles We Want to Bring In — And Why They Belong Inside Existing Specialists

When users install agency-agents into Claude Code, Cursor, Gemini CLI, OpenCode, or Antigravity, they rarely invoke a separate "methodology critic" agent mid-task — they invoke the domain specialist directly (engineering-rag-pipeline-engineer, security-secrets-credential-engineer, engineering-code-reviewer, engineering-sre).

Right now, many agents excel at happy-path architecture and framework syntax, but lack concrete guardrails against two classes of real-world production failures:

1. Pragmatic Systems Rigor (inspired by The Pragmatic Programmer):

   • Design by Contract (DbC): Explicit preconditions, postconditions, and semantic invariants (e.g., zero-sum double-entry conservation in payments/data pipelines) rather than trusting implicit caller behavior.
   • Crash Early & Deterministic Resource Balancing (LIFO): Stopping corrupted state before it poisons persistent storage, and guaranteeing deterministic cleanup of file descriptors, DB locks, PTY sessions, and dynamic credential leases on timeout/error paths.
   • State Coverage over Shallow Line Coverage: Testing boundary state transitions, concurrent race barriers (TOCTOU), and mutation survival rather than happy-path-only assertions.

2. Stochastic & Agentic Failure Modes (inspired by production AI engineering / Competente o Suficiente):

   • Context Rot & Retrieval Abstention: Preventing RAG pipelines from stuffing 20+ chunks into context windows (Lost-in-the-Middle attention degradation), enforcing exact lexical/metadata routing before fuzzy cosine similarity, and short-circuiting with ABSTAIN_INSUFFICIENT_EVIDENCE when calibrated reranker scores fall below threshold.
   • Zero Ambient Authority & Indirect Prompt Injection Quarantine: Ensuring security, email, and MCP agents never inherit raw host credentials (.env, ~/.aws, ~/.ssh) or execute privileged tool calls triggered by untrusted external text.
   • Deterministic Harnesses over Open-Ended Loops: Replacing free-form multi-agent chat (where reliability decays exponentially as $p^n$, e.g., $0.95^{20} \approx 35%$) with typed state machines, compiler gates, and immutable holdout evaluators.

Why not put this in 1–2 dedicated agents?
We already have specialized-master-plan-architect and security-ai-generated-code-auditor for dedicated planning and auditing. However, if engineering-rag-pipeline-engineer doesn't enforce retrieval abstention and prompt-injection quarantine while designing the RAG pipeline, or if security-secrets-credential-engineer doesn't block coding agents from reading .env files while setting up secret management, the vulnerability ships before a secondary auditor is ever called. Each specialist only needs the 3–4 failure modes native to its exact domain.

---

2. Which Agents Gain the Most — And Which We'd Leave Alone

Looking critically at our original 86-agent scope, we cast the net too wide. We do not need to modify 86 agents upstream.

A. What We Would Leave Alone (~225+ agents across the catalog, including ~20–25 from our original branches)

• All ~197 agents outside our branches across design, sales, paid-media, finance, healthcare, gis, and regional/platform marketing.
• ~17–25 narrative, coaching, and general PM/product/marketing agents from our branches where formal DbC contracts or deterministic harnesses feel heavy-handed for their core job — for example:
  • marketing/marketing-book-co-author.md
  • specialized/personal-growth-mentor.md
  • specialized/zk-steward.md
  • project-management/project-management-meeting-notes-specialist.md
  • project-management/project-management-project-shepherd.md
  • product/product-feedback-synthesizer.md
  • product/product-sprint-prioritizer.md
  • research/research-synthesist.md
  • Any agent already at 400–550+ lines where you feel current weight is already optimal.

B. Which Agents Gain the Most (Our Proposed Priority Tiers)

1. Wave 1 — Under-Specified Foundational Stubs (43–150 lines currently):
   Several of the most frequently installed foundational agents are still early 43–90 line skeletons with minimal or no runnable deliverables:
   • engineering/engineering-code-reviewer.md (59 lines on main → 110 lines)
   • engineering/engineering-sre.md (74 lines on main → 103 lines)
   • engineering/engineering-git-workflow-master.md (79 lines on main → 123 lines)
   • engineering/engineering-software-architect.md (87 lines on main → 117 lines)
   • specialized/specialized-document-generator.md (43 lines on main → 215 lines)
   • spatial-computing/terminal-integration-specialist.md (57 lines on main → 286 lines)
2. Wave 2 — AI, RAG, Multi-Agent & Data Specialists:
   Where stochastic failure modes (context rot, indirect prompt injection, compound error decay $p^n$, look-ahead data leakage) directly break production systems:
   • engineering-rag-pipeline-engineer, engineering-ai-engineer, engineering-llm-post-training-engineer, engineering-multi-agent-systems-architect, engineering-email-intelligence-engineer, engineering-data-engineer, specialized-model-qa, specialized-mcp-builder.
3. Wave 3 — Security, Testing & High-Consequence Reliability Specialists:
   Where agentic execution, credential exposure, and concurrency bugs create severe blast radius:
   • security-secrets-credential-engineer, specialized/agentic-identity-trust, security-appsec-engineer, engineering-database-reliability-engineer, engineering-payments-billing-engineer, engineering-rust-refactoring-specialist, testing-test-automation-engineer, testing-api-tester, testing-test-results-analyzer.

---

3. Before-and-After on Real Tasks (Two Concrete Examples)

Example A: engineering/engineering-code-reviewer.md (59 lines → 110 lines, +51 net lines)

• Before (on main): A 59-line high-level checklist ("Check for security vulnerabilities, race conditions, missing error handling, code duplication"). On a real PR review task, it tends to produce generic textbook reminders.
• After: Adds a concrete Pragmatic Code Review Matrix that catches subtle production bugs AI-assisted PRs routinely introduce:
  1. The 4 DRY Categories: Distinguishes Inadvertent Duplication (storing both lineItems and a mutable cachedOrderTotal column that drifts out of sync) from Imposed, Impatient, or Interdeveloper duplication.
  2. Side-Effect Assertions & Catch-and-Release: Flags assertions that mutate state (assert(iter.next() != null)) and try/catch blocks that swallow exceptions and let corrupted state continue running (Dead Programs Tell No Lies).
  3. Canonical Lock Ordering & LIFO Resource Symmetry: Detects AB/BA deadlocks when locking multiple records out of primary-key order and leaked file descriptors/transactions on early return paths.
  4. Stroop Effect Naming: Flags contradictory identifiers (e.g., disableRetry = false, or accountList holding a Map).

Example B: engineering/engineering-rag-pipeline-engineer.md (365 lines → 463 lines, +98 net lines)

• Before (on main): Covers chunking, hybrid BM25 + dense retrieval, and RRF reranking well, but passes top-$K$ chunks straight to the LLM without a token ceiling, abstention threshold, or prompt-injection filter.
• After: Adds a runnable assemble_contract_verified_context() gate between retrieval and generation that changes how the agent builds RAG pipelines on real tasks:
  1. Abstention Contract (ABSTAIN_INSUFFICIENT_EVIDENCE): If the best calibrated reranker score falls below threshold (< 0.35), it short-circuits before calling the LLM instead of letting the model improvise from parametric memory.
  2. Anti-Context-Rot Packing: Enforces a hard token ceiling (2,800 tokens) and places the highest-scoring evidence at the start and end of the context window to mitigate Lost-in-the-Middle attention decay.
  3. Untrusted Chunk Quarantine: Scans retrieved chunks for indirect prompt-injection directives (ignore previous instructions, <|im_start|>, etc.) and quarantines poisoned documents before they reach the generator.

---

4. Keeping Each Agent's Voice, Scope, and Length Bounded

To ensure every agent keeps its own personality and scope:

1. Zero Boilerplate Copy-Paste (Verified by check-agent-originality.sh):
   Every addition uses only the idioms, tools, and runnable code of that specific specialist (PySpark in Data Engineer, Swift/PTY in Terminal Specialist, bubblewrap/Vault in Secrets Engineer, Playwright in Test Automation, Cargo/Miri in Rust Specialist). Across all 283 agents in the repo, scripts/check-agent-originality.sh passes with 0 warnings and 0 failures (overlap across our modified agents is 0.0%–3.4%, well below the 20% warning threshold).
2. Existing Persona & Upstream Fixes Preserved 100%:
   We preserve each agent's frontmatter, personality, vibe, and existing examples — including 100% of the recent upstream fixes from #1132–#1177 (such as @rudycelekli's fixes in engineering-devops-automator.md, engineering-data-engineer.md, engineering-frontend-developer.md, support-analytics-reporter.md, etc.).
3. Bounded Length Growth:
   • Across the set, the median net addition is +69 lines per agent.
   • Short 43–87 line stubs grow to ~105–215 lines so they have real code deliverables on par with the rest of the catalog.

---

5. Rebased Division Branches Ready for Granular Review (1 Skill = 1 Commit)

To make reviewing as easy as possible once we agree on scope, we have already rebased everything onto current main (b37b5bb) and organized the work into division branches where every single skill upgrade is 1 isolated, atomic commit (1 commit = 1 file):

• Wave 1 — Core Engineering Foundations (11 commits, 1 per agent): augustoheiss:feat/div-engineering-foundations
• Wave 2A — AI, RAG, Post-Training & Data Engineering (11 commits, 1 per agent): augustoheiss:feat/div-engineering-ai-and-data
• Wave 2B — Backend, DBRE, Rust, DevOps & Systems Engineering (15 commits, 1 per agent): augustoheiss:feat/div-engineering-systems
• Wave 3A — Security Division (9 commits, 1 per agent): augustoheiss:feat/div-security
• Wave 3B — Testing Division (9 commits, 1 per agent): augustoheiss:feat/div-testing
• Wave 3C — Core Technical Specialized Agents (8 commits, 1 per agent): augustoheiss:feat/div-specialized-core
• Wave 4 — Support (2), Game Dev (3), Spatial Computing (1): feat/div-support, feat/div-game-development, feat/div-spatial-computing

Please let us know which division or subset of agents you'd like us to send first (for example, starting with a small PR of the short foundational stubs in feat/div-engineering-foundations), and which ones you'd prefer to skip!

[augustoheiss]

Complete Commit-by-Commit Catalog (1 Skill = 1 Atomic Commit on main @ b37b5bb)

We have opened Wave 1 (feat/div-engineering-foundations) as #1241 so the 11 foundational engineering agents can be reviewed commit-by-commit.

Below is the full index of all 86 atomic commits across the 10 division branches on augustoheiss/agency-agents, with each commit touching exactly 1 agent file and describing its specific upgrade:

Wave 1 — Core Engineering Foundations (11 Commits — Opened as #1241)

Branch: augustoheiss:feat/div-engineering-foundations

#	Atomic Commit	Skill File	Commit Description
1	2d65f8c	engineering/engineering-code-reviewer.md	feat(engineering): add 4-DRY audit matrix, Law of Demeter checks, and LIFO resource symmetry to Code Reviewer
2	a86899f	engineering/engineering-sre.md	feat(engineering): add crash-early supervisor patterns, resource leak canaries, and diagnostic windows to SRE
3	bed0acb	engineering/engineering-git-workflow-master.md	feat(engineering): add automated git-bisect fault isolation and 4-stage clean build contracts to Git Workflow Master
4	04acbec	engineering/engineering-software-architect.md	feat(engineering): add orthogonality audits, Design by Contract boundaries, and blackboard coordination to Software Architect
5	601080a	engineering/engineering-senior-developer.md	feat(engineering): add deliberate programming guards and testable GPU/state inspection pins to Senior Developer
6	5f0e863	engineering/engineering-minimal-change-engineer.md	feat(engineering): add two-hat commit separation and binary-chop debugging discipline to Minimal Change Engineer
7	eae9770	engineering/engineering-developer-tooling-engineer.md	feat(engineering): add plain-text pipeline contracts and multi-target active code generation to Developer Tooling Engineer
8	fdd608a	engineering/engineering-api-platform-engineer.md	feat(engineering): add API boundary Design by Contract and executable schema DRY patterns to API Platform Engineer
9	be1a52e	engineering/engineering-database-optimizer.md	feat(engineering): add canonical lock ordering and empirical scaling benchmarks to Database Optimizer
10	859adb8	engineering/engineering-orgscript-engineer.md	feat(engineering): add canonical EBNF grammar and multi-target AST compilation to OrgScript Engineer
11	c3f78fc	engineering/engineering-rapid-prototyper.md	feat(engineering): add disposable prototype vs evolutionary tracer-bullet contracts to Rapid Prototyper

Wave 2A — AI, RAG, Post-Training & Data Engineering (11 Commits)

Branch: augustoheiss:feat/div-engineering-ai-and-data

#	Atomic Commit	Skill File	Commit Description
1	3dad788	engineering/engineering-ai-engineer.md	feat(engineering): add provider reversibility ports and contract-enforced inference runners to AI Engineer
2	c96e071	engineering/engineering-prompt-engineer.md	feat(engineering): add orthogonal prompt specifications, abstention contracts, and instruction ablation to Prompt Engineer
3	81f2da7	engineering/engineering-data-engineer.md	feat(engineering): add Write-Audit-Publish conservation contracts and dead-letter quarantine to Data Engineer
4	26e97aa	engineering/engineering-rag-pipeline-engineer.md	feat(engineering): add retrieval grounding contracts, token budget ceilings, and injection quarantine to RAG Pipeline Engineer
5	9826bce	engineering/engineering-llm-post-training-engineer.md	feat(engineering): add retrieval-vs-SFT gates, holdout decontamination, and quantization parity checks to LLM Post-Training Engineer
6	8f530ed	engineering/engineering-search-relevance-engineer.md	feat(engineering): add L0 exact-match guardrails and walk-forward temporal LTR splits to Search Relevance Engineer
7	3894f7a	engineering/engineering-knowledge-graph-engineer.md	feat(engineering): add bitemporal fact validity and pre-commit SHACL harnesses to Knowledge Graph Engineer
8	9105268	engineering/engineering-ai-data-remediation-engineer.md	feat(engineering): add deterministic postcondition validators and append-only remediation ledgers to AI Data Remediation Engineer
9	80ab2ef	engineering/engineering-email-intelligence-engineer.md	feat(engineering): add untrusted payload quarantine and deterministic policy enforcement gates to Email Intelligence Engineer
10	a56796d	engineering/engineering-voice-ai-integration-engineer.md	feat(engineering): add multi-stage Whisper silence hallucination filters and LIFO buffer cleanup to Voice AI Integration Engineer
11	d52727e	engineering/engineering-multi-agent-systems-architect.md	feat(engineering): add compound-error reliability bounds and typed blackboard state machines to Multi-Agent Systems Architect

Wave 2B — Backend, DBRE, Rust, DevOps & Systems Engineering (15 Commits)

Branch: augustoheiss:feat/div-engineering-systems

#	Atomic Commit	Skill File	Commit Description
1	a0af8a7	engineering/engineering-backend-architect.md	feat(engineering): add complexity justification gates, lexicographical lock ordering, and transactional outbox to Backend Architect
2	33ff457	engineering/engineering-database-reliability-engineer.md	feat(engineering): add mandatory lock/statement timeouts and hardware-physics sizing to Database Reliability Engineer
3	d357fe3	engineering/engineering-rust-refactoring-specialist.md	feat(engineering): add golden-parity polyglot migration suites and Miri verification harnesses to Rust Refactoring Specialist
4	c2cd188	engineering/engineering-devops-automator.md	feat(engineering): add 40-char SHA action pinning, lean hosting TCO gates, and atomic rollback scripts to DevOps Automator
5	af44c6e	engineering/engineering-autonomous-optimization-architect.md	feat(engineering): add SHA-256 immutable holdout verification and Occam complexity gates to Autonomous Optimization Architect
6	21207a4	engineering/engineering-codebase-onboarding-engineer.md	feat(engineering): add dual ONBOARDING/AGENTS contracts and executable clean-clone verification to Codebase Onboarding Engineer
7	29298c6	engineering/engineering-technical-writer.md	feat(engineering): add Vale anti-slop rules and CI markdown doctest verification to Technical Writer
8	c993810	engineering/engineering-frontend-developer.md	feat(engineering): add zero-bloat dependency gates and discriminated-union state machines to Frontend Developer
9	69b9dc9	engineering/engineering-payments-billing-engineer.md	feat(engineering): add append-only double-entry ledger invariants and automated refund circuit breakers to Payments Billing Engineer
10	526043a	engineering/engineering-finops-engineer.md	feat(engineering): add agent token budget guards and cloud-API vs dedicated-GPU break-even models to FinOps Engineer
11	d847a1f	engineering/engineering-platform-engineer.md	feat(engineering): add CLI-first dry-run contracts and Provision-equals-Teardown lifecycle checks to Platform Engineer
12	7dcde74	engineering/engineering-embedded-firmware-engineer.md	feat(engineering): add host-side ASan/UBSan bus simulation and static allocation guards to Embedded Firmware Engineer
13	1dcfd13	engineering/engineering-i18n-engineer.md	feat(engineering): add NFC boundary normalization, grapheme-safe truncation, and ICU AST parity checks to i18n Engineer
14	6089469	engineering/engineering-universal-document-compiler.md	feat(engineering): add byte-for-byte CST span conservation and Zero-Trust PDF sanitization to Universal Document Compiler
15	af732f2	engineering/engineering-privacy-engineer.md	feat(engineering): add cascading subject erasure across SQL, vector indices, and KMS stores to Privacy Engineer

Wave 3A — Security Division (9 Commits)

Branch: augustoheiss:feat/div-security

#	Atomic Commit	Skill File	Commit Description
1	1142502	security/security-ai-generated-code-auditor.md	feat(security): add AI code CWE detectors, catch-and-release traps, and resource leak audits to AI-Generated Code Security Auditor
2	794e9b0	security/security-secrets-credential-engineer.md	feat(security): add opaque egress credential brokers and bubblewrap agent secret blackouts to Secrets & Credential Engineer
3	e9f8a0f	security/security-architect.md	feat(security): add stochastic trust boundary modeling and gVisor tool-execution isolation to Security Architect
4	6558928	security/security-appsec-engineer.md	feat(security): add O_NOFOLLOW symlink traversal guards, atomic TOCTOU patterns, and slopsquatting checks to AppSec Engineer
5	3c7414e	security/security-cloud-security-architect.md	feat(security): add GPU/agent subnet egress firewalling and IMDSv2 hop-limit policies to Cloud Security Architect
6	fae3164	security/security-penetration-tester.md	feat(security): add indirect prompt-injection exploit chains and barrier-synchronized TOCTOU harnesses to Penetration Tester
7	013ff78	security/security-incident-responder.md	feat(security): add SIGSTOP volatile-memory preservation and dynamic lease revocation runbooks to Incident Responder
8	75143cf	security/security-threat-detection-engineer.md	feat(security): add MCP subprocess credential-access Sigma rules and PPV base-rate calibration to Threat Detection Engineer
9	f0f2453	security/security-compliance-auditor.md	feat(security): add 100%-population Compliance-as-Code collectors and AI governance controls to Compliance Auditor

Wave 3B — Testing Division (9 Commits)

Branch: augustoheiss:feat/div-testing

#	Atomic Commit	Skill File	Commit Description
1	47bd0f0	testing/testing-test-automation-engineer.md	feat(testing): add state-coverage boundary matrices and mutation saboteur harnesses to Test Automation Engineer
2	00f8dc6	testing/testing-evidence-collector.md	feat(testing): add wild real-world boundary payloads and git-bisect regression isolation to Evidence Collector
3	3e859d7	testing/testing-reality-checker.md	feat(testing): add 6-pillar production certification gates and automated broken-window audits to Reality Checker
4	27f4979	testing/testing-performance-benchmarker.md	feat(testing): add empirical Big-O doubling-ratio curves and Little's Law capacity models to Performance Benchmarker
5	2fee80a	testing/testing-api-tester.md	feat(testing): add 20-way race-barrier idempotency tests and Schemathesis property fuzzing to API Tester
6	34bf63c	testing/testing-test-results-analyzer.md	feat(testing): add AST test-theater detection and mutation kill-rate gates to Test Results Analyzer
7	108dade	testing/testing-tool-evaluator.md	feat(testing): add unsubsidized 3x TCO stress modeling and plain-text exit cost audits to Tool Evaluator
8	f977971	testing/testing-workflow-optimizer.md	feat(testing): add deterministic-first automation hierarchy and compound reliability math to Workflow Optimizer
9	94c6b18	testing/testing-accessibility-auditor.md	feat(testing): add AI-generated UI anti-pattern checks and streaming aria-live debounce audits to Accessibility Auditor

Wave 3C — Core Technical Specialized Agents (8 Commits)

Branch: augustoheiss:feat/div-specialized-core

#	Atomic Commit	Skill File	Commit Description
1	87f4f45	specialized/specialized-master-plan-architect.md	feat(specialized): add orthogonality audits, Design by Contract tables, tracer-bullet matrices, and PERT estimation to Master Plan Architect
2	5476859	specialized/specialized-document-generator.md	feat(specialized): add single-source plain-text manifest compilation across XLSX, DOCX, and PPTX to Document Generator
3	e789cc9	specialized/specialized-mcp-builder.md	feat(specialized): add Law-of-Demeter tool boundaries, subpath containment, and AsyncExitStack lifecycle management to MCP Builder
4	81a9676	specialized/specialized-codebase-archaeologist.md	feat(specialized): add 4-DRY forensic taxonomy and architectural entropy heatmaps to Codebase Archaeologist
5	66d59de	specialized/agents-orchestrator.md	feat(specialized): add orthogonal subagent partitioning and blackboard state coordination to Agents Orchestrator
6	1fc3a6d	specialized/agentic-identity-trust.md	feat(specialized): add Zero Ambient Authority spawn sanitization and external policy enforcement gates to Agentic Identity & Trust Architect
7	3b6a4ab	specialized/specialized-model-qa.md	feat(specialized): add out-of-sample naive baseline gates, walk-forward leakage checks, and Simpson's paradox stratification to Model QA Specialist
8	dc1919e	specialized/lsp-index-engineer.md	feat(specialized): add sub-50ms semantic context capsules and UTF-16/UTF-8 byte-offset parity to LSP Index Engineer

Wave 4A — Support Division (2 Commits)

Branch: augustoheiss:feat/div-support

#	Atomic Commit	Skill File	Commit Description
1	aff94ff	support/support-infrastructure-maintainer.md	feat(support): add hardware-physics USE diagnostics, pre-restart forensic capture, and quorum-checked dry-run scripts to Infrastructure Maintainer
2	3762167	support/support-analytics-reporter.md	feat(support): add Simpson's paradox subgroup detectors and walk-forward P10/P50/P90 interval validation to Analytics Reporter

Wave 4B — Game Development Division (3 Commits)

Branch: augustoheiss:feat/div-game-development

#	Atomic Commit	Skill File	Commit Description
1	146b4ad	game-development/unity/unity-architect.md	feat(game-development): add deterministic simulation decoupling, ScriptableObject play-mode reset, and zero-alloc update guards to Unity Architect
2	6b2f798	game-development/godot/godot-gameplay-scripter.md	feat(game-development): add headless CI static verification and node-based finite state machines to Godot Gameplay Scripter
3	f9fe5fb	game-development/unreal-engine/unreal-systems-engineer.md	feat(game-development): add UHT API verification guards, assertive checkf contracts, and VRAM pool budgets to Unreal Systems Engineer

Wave 4C — Spatial Computing Division (1 Commit)

Branch: augustoheiss:feat/div-spatial-computing

#	Atomic Commit	Skill File	Commit Description
1	97370ca	spatial-computing/terminal-integration-specialist.md	feat(spatial-computing): add ManagedPTYSession LIFO descriptor cleanup and 120Hz frame-coalesced terminal feeding to Terminal Integration Specialist

Wave 5 — Product, PM, Marketing, Academic/Research & Governance (17 Commits — Optional / Hold)

Branch: augustoheiss:feat/div-product-and-pm

#	Atomic Commit	Skill File	Commit Description
1	4557d16	product/product-dx-engineer.md	feat(product): add dual human/agent CLI ergonomics, JSON stdout envelopes, and dry-run contracts to DX Engineer
2	6035449	product/product-manager.md	feat(product): add two-step Socratic discovery, deterministic baseline gates, and P50/P95/P99 token margin tables to Product Manager
3	83a8abf	product/product-feedback-synthesizer.md	feat(product): add invariant requirement vs mutable policy excavation to Feedback Synthesizer
4	b637b68	product/product-sprint-prioritizer.md	feat(product): add vertical tracer-bullet slicing and protected entropy-reduction budgets to Sprint Prioritizer
5	e14339c	project-management/project-manager-senior.md	feat(project-management): add 3-point PERT estimation and horizon-calibrated precision units to Senior Project Manager
6	1fd2d39	project-management/project-management-meeting-notes-specialist.md	feat(project-management): add project glossary delta extraction and ceremony-automation audits to Meeting Notes Specialist
7	330ddbd	project-management/project-management-project-shepherd.md	feat(project-management): add 90%-done syndrome defenses and 3-point PERT schedule buffers to Project Shepherd
8	87610a1	project-management/project-management-jira-workflow-steward.md	feat(project-management): add anti-megadiff PR size gates and stacked-PR slicing rules to Jira Workflow Steward
9	bcfbcfd	project-management/project-management-experiment-tracker.md	feat(project-management): add SHA-256 pre-registration locks and chi-square Sample-Ratio Mismatch circuit breakers to Experiment Tracker
10	4f4e7ad	marketing/marketing-aeo-foundations.md	feat(marketing): add discovery-surface route hygiene and build-time DRY llms.txt compilation to AEO Foundations Architect
11	56239b6	marketing/marketing-agentic-search-optimizer.md	feat(marketing): add two-phase human confirmation gates and SHA-256 idempotency headers to Agentic Search Optimizer
12	7417fb1	marketing/marketing-book-co-author.md	feat(marketing): add Socratic voice-preservation protocols and cross-chapter factual continuity ledgers to Book Co-Author
13	3547df6	academic/academic-statistician.md	feat(academic): add fold-isolated TimeSeriesSplit pipelines and ADF/KPSS stationarity triage to Statistician
14	88ff86a	research/research-synthesist.md	feat(research): add ghost-reference verification gates and primary-source citation graph deduplication to Research Synthesist
15	031c13a	specialized/automation-governance-architect.md	feat(specialized): add deterministic-first workflow gates, compound reliability audits, and blast-radius ceilings to Automation Governance Architect
16	5ca316a	specialized/personal-growth-mentor.md	feat(specialized): add anti-cognitive-atrophy struggle protocols and 80/20 durable fundamentals portfolios to Personal Growth Mentor
17	d873de7	specialized/zk-steward.md	feat(specialized): add typed epistemic link contracts and bitemporal fact supersession to Zettelkasten Steward

[augustoheiss]

Summary of Division-Scoped Pull Requests (#1241 – #1250)

All 10 division-scoped branches (1 commit = 1 skill, rebased cleanly onto origin/main @ b37b5bb) have been opened as separate, self-contained PRs with full commit-by-commit documentation tables so each division (or any individual agent commit inside a division) can be reviewed, cherry-picked, or closed independently:

Wave	PR	Division / Scope	Atomic Commits (1 per Skill)	Branch
Wave 1	#1241	Engineering — Foundational Stubs & Core Architecture (code-reviewer, sre, git-workflow-master, software-architect, senior-developer, etc.)	11 commits	feat/div-engineering-foundations
Wave 2A	#1242	Engineering — AI, RAG, Post-Training, Graph & Data (rag-pipeline-engineer, ai-engineer, llm-post-training-engineer, multi-agent-systems-architect, etc.)	11 commits	feat/div-engineering-ai-and-data
Wave 2B	#1243	Engineering — Backend, DBRE, Rust, DevOps, Frontend, Payments & Systems	15 commits	feat/div-engineering-systems
Wave 3A	#1244	Security Division (secrets-credential-engineer, architect, appsec-engineer, cloud-security-architect, etc.)	9 commits	feat/div-security
Wave 3B	#1245	Testing Division (test-automation-engineer, api-tester, test-results-analyzer, tool-evaluator, etc.)	9 commits	feat/div-testing
Wave 3C	#1246	Specialized — Core Technical Agents (master-plan-architect, document-generator, mcp-builder, agentic-identity-trust, specialized-model-qa, lsp-index-engineer, etc.)	8 commits	feat/div-specialized-core
Wave 4A	#1247	Support Division (infrastructure-maintainer, analytics-reporter)	2 commits	feat/div-support
Wave 4B	#1248	Game Development Division (unity-architect, godot-gameplay-scripter, unreal-systems-engineer)	3 commits	feat/div-game-development
Wave 4C	#1249	Spatial Computing Division (terminal-integration-specialist)	1 commit	feat/div-spatial-computing
Wave 5	#1250	Product, PM, Marketing, Academic/Research & Governance	17 commits	feat/div-product-and-pm