Features

Chief Architect for system integrity, scalability, and alignment with architectural patterns. Leads context gathering, solution exploration, and user consultation.

Simplifies and refines code for clarity, consistency, and maintainability while preserving all functionality. Focuses on recently modified code unless instructed otherwise.

Runs the post-QA hardening workflow (7 steps): re-validate vs PRD/task, simplify code, check references/coverage/lint/tests, produce compliance report.

Senior Developer implementing features with TDD and high code quality. Follows project patterns and style guide.

DevOps Engineer owning the build-ship-run pipeline from code commit to running in production.

Plugin-backed subagent inference layer that classifies an unknown Datarim pane line into a slash-command via a multi-backend AI CLI chain (claude → codex → cursor). Fail-closed; threshold gating lives in the caller.

Content Editor for editorial review, fact verification, AI pattern removal, and publication-ready quality. Works with articles, blog posts, social media, research papers, and documentation.

Knowledge Base Librarian for organizing, indexing, cross-referencing, and maintaining the datarim/ directory. Runs ingest, lint, and consolidation workflows to keep the knowledge base healthy and navigable.

Audit and improve the Datarim framework: detect bloat, duplicates, oversized files, weak descriptions, and selective-loading opportunities.

Adversarial reviewer subagent for /dr-verify Layer 2/3 cross-Claude-family fallback. Reviews PRD/plan/code in clean isolated context. Findings-only, read-only tools.

Lead Project Manager for backlog management, detailed design, implementation plans, and complexity levels.

Investigates external context — library versions, best practices, CVEs, compatibility — before planning begins and when implementation hits unknowns.

QA and Security Lead for code reviews, security compliance, and Definition of Done validation.

Security Analyst identifying, assessing, and mitigating security risks throughout the development lifecycle.

Skill Creator agent for researching, designing, and generating new skills, agents, and commands. Analyzes best practices, audits existing framework components, and creates or updates artifacts in the correct location.

Site Reliability Engineer ensuring systems are reliable, observable, and recoverable in production.

Strategic Advisor evaluating whether a task is worth building and proposing the most efficient path.

Platform QA agent for verifying changes across any project. Auto-detects test runners, supports Docker-aware execution, API smoke tests, and structured result reporting.

Content Writer for creating articles, blog posts, social media content, research papers, technical documentation, and any structured written output. Focuses on clear, engaging, audience-appropriate writing.

Create or update skills, agents, and commands

Archive completed task with mandatory reflection and evolution proposals

Subagent orchestrator — drives a task to a passing /dr-compliance + reflection (not the final archive) by spawning per-stage agents and summarising their results, with the FB-rules and Question Suppression Ladder on by default.

Adaptive post-QA hardening with type-specific checklists

Resume from session handoff artefact in a clean context window

Explore architectural and design decisions for complex features

TDD implementation with gap discovery — auto-spawns researcher on unknowns

Diagnose and migrate operational files to thin-index schema

Knowledge base maintenance, organization, and consolidation

Editorial review with fact-checking and AI pattern removal

Explain an existing result or unfamiliar term again, without executing work or changing task status.

Show the complete Datarim command reference

Initialize a new task, resume from backlog, or scaffold a new project

Resume work on current task from last checkpoint

Audit and optimize the Datarim framework

Self-driving Datarim pipeline with snapshot-first resume, fail-closed routing, and operator-confirmed learned rules.

Create a detailed implementation plan with security analysis

Manage opt-in plugins: list, enable, disable, sync, doctor

Generate PRD with structured research phase (Phase 1.3) for L2+ tasks

Prepare platform-adapted publish payloads for approved content (does not send them)

Multi-layer quality verification across PRD, design, plan, and code

Fast-lane for trivial fixes or quick lookups — skips the heavy pipeline

Save current session state to disk before context is destroyed

Check current task status, progress, and backlog summary

Standalone tri-layer self-verification: deterministic floor, runtime-owned peer review, and native read-only dispatch

Build a task specification through a resumable, one-question-at-a-time interview

Create written content with structured writing workflow

Fact-check a file before publication

Remove AI writing patterns from text

Forces an adversarial mindset on a plan, PRD, design, or code change — break it rather than bless it. Loaded at the /dr-plan checkpoint and /dr-qa.

Five pillars of AI-assisted development — decomposition, TDD, architecture-first, focused work, context. Method size limits, DoD, stubbing.

Canonical Datarim conventions reference for native agents generating or editing project-local artifacts — Stage Header, byte-exact YAML frontmatter, append-log Q&A, expectations checklist, snapshot frontmatter, PRD↔archive mirror, Diátaxis taxonomy. Loaded by /dr-write and /dr-archive.

Question Suppression Ladder + L1 Inline Resolution Rule + Hard-gated Action Boundary. Activated by DATARIM_AUTO_MODE=1 + a per-task autonomous-mode marker.

MUST use this before any creative work — features, components, functionality, behavior changes. Explores user intent and design before implementation.

An open format for @cc directives colocating structured assumptions and requirements with code — specification, attention, and verification for both humans and agents, enforced by the cc-check CLI.

Post-QA hardening — detects task type (code, docs, research, legal, content, infra) and applies the matching verification checklist before archiving.

Assemble a panel of Datarim agents for multi-perspective analysis of decisions. Use for /dr-design stage (L3-4) or critical decisions in /dr-plan.

Default-off orchestrator contract for checkpoint-before-reset context compaction/clearing at deterministic pressure thresholds — snapshot-first continuity across Claude Code and Codex.

Layered timeout defense for cron-orchestrated agents calling external APIs (LLM CLI, HTTP, subprocess) — nested tiers, anti-patterns, headroom.

Canonical CTA "Next Step" block format for every /dr-* command and pipeline agent. Single source of truth — load when generating slash-command output.

Trace a user-facing requirement from its verbatim source through implementation, evidence, deployment, live proof, and explicit disposition.

Schema spec for thin-index operational files: canonical regex, YAML frontmatter contract, 8-pass migration semantics (Pass 0–7, incl. archive-section enforcement), data-loss safety contract. Loaded on demand by /dr-doctor and /dr-init self-heal.

Core Datarim rules. Load this entry first, then only the fragment needed for paths, storage, numbering, backlog, routing, or archive behavior.

Mandate for documentation taxonomy across Datarim-managed repos and product sites: four orthogonal categories (tutorials, how-to, reference, explanation), closed mapping table, exemption list, anti-pattern definitions. Loaded by /dr-init, /dr-optimize audit, /dr-archive surface verification.

Structured requirements discovery through focused one-question-at-a-time interviews with proposed answers. Use in /dr-prd stage or before /dr-init.

Use when facing 2+ independent tasks that can be worked on without shared state or sequential dependencies

Detect and resolve task-ID collisions in the TOCTOU window between /dr-init reservation and /dr-archive completion; reserve IDs with the atomic allocator.

Consumer contract for stage snapshots — /dr-next and /dr-orchestrate read datarim/snapshots/{TASK-ID}.snapshot.md first, emit replay-prompt with CTA + bilingual autonomy reminder + done before: body.

Knowledge base maintenance — organize, deduplicate, cross-reference, and consolidate the datarim/ directory. Flags contradictions, archives stale content.

Enumerates boundary, degenerate, and failure inputs an artifact does not visibly handle. Sibling of adversarial-review; loaded at /dr-plan and /dr-qa.

Rules for proposing and applying framework improvements. Covers growth (new components) and maintenance (pruning, merging). Human approval required.

Use when you have a written implementation plan to execute in a separate session with review checkpoints

Operator wishlist artefact in plain markdown — wish_id slug, status history, current status, override semantics. Written at the requirements step, verified at the quality-review and hardening steps.

Fact-check articles and posts before publication. Extracts claims, verifies against sources, improves text while preserving style.

Pre-flight checklist + ignore patterns for file-sync (Syncthing/rclone/rsync/Dropbox/iCloud) — protects git working trees, virtualenvs, and build artifacts.

Use when implementation is complete and all tests pass. Decides how to integrate the work via merge, PR, or cleanup with structured options.

Fleet starter skills across five complexity tiers (L1–L5) — each injects a context budget matched to the task level, from a single-step command to self-managed multi-subtask coordination.

Minimal starter context for one elementary, single-step command.

Compact starter context for a few templated steps within an allowed command list.

On-demand retrieval and explicit choice for work with variability and alternatives.

Structured decomposition and per-subtask retrieval for complex multi-step work.

Self-managed coordination for complex work, with bounded authority and knowledge-base prefetch.

Frontend UI checklist — CSS specificity, dark/light themes, visual testing, mobile responsiveness, i18n parity. Apply when editing HTML/CSS.

Surface hard-coded stub literals (pending-integration, not-implemented, stub) in health/status controllers at /dr-do, before /dr-qa wish gating.

Explain what the user can now do, which requirements were verified, and what remains open. Works with Datarim or as a standalone skill.

Compatible four-section recap for QA, compliance, and archive, governed by Human Outcome Reporting. Completeness before brevity.

Remove AI writing patterns and formatting artifacts from text. Preserves author voice. Use for articles, posts, content that must not look AI-generated.

Author prompts for image-generation tools — covers, thumbnails, post visuals, illustrations, infographics, logos; intake→spec→prompt→verify loop.

Immutability contract for all pipeline stages: artefact freeze, V-AC parity, non-code parity, anti-tautological rule, and return-to-source transition.

SSH batch execution, ping matrices, health checks for your servers. Use when performing infrastructure operations across multiple servers.

Verbatim operator brief artefact contract — frontmatter + append-log + Q&A round-trip auto-append + mandatory read by every pipeline command. Source of truth for operator intent across the task lifecycle.

Allowlist/blocklist for network bind targets (compose ports, redis bind, postgres listen_addresses, systemd ListenStream); load before any port change.

Probe the running nginx version before writing config, then map directive syntax to that version. Load in /dr-plan for any nginx-touching task.

Optimization patterns (lazy loading, caching, batching), database and frontend performance. Use when designing or reviewing for performance.

Exists-check for file/script/path references in /dr-plan output — flags missing or deprecated tooling and file references before they reach /dr-do.

Browser-based frontend quality check — resolution chain across CLI, MCP, and operator-specified browser; default headless mode, optional visible mode, hard-fail strict visible mode.

Pre-archive gate diffing the on-host env file against the repo template when a deploy changed defaults — catches «prod on stale .env».

Deploy-class prod-readiness gate — a read-only test↔prod runner symmetry probe that blocks merge-proposal and archive until production is verified. Hybrid: deterministic with a project contract, agent-checklist without one.

Project scaffolding — creates AGENTS.md, documentation/, and datarim/ for new or existing projects.

Technical rules for publishing content to social media and websites — platform limits, formatting, API patterns, OG tags, multi-platform workflow.

Use when receiving review feedback before implementing suggestions. Requires technical verification, not performative agreement or blind implementation.

Review-phase workflow — lessons learned, evolution proposals (Class A/B gate), health-metrics check. Runs at /dr-compliance; /dr-archive Step 0.5 regenerates it only if missing or stale.

Consumer-side recipe for verifying Datarim releases — sha256 → cosign verify-blob → gh attestation verify. Load on install/update from GitHub Release.

Use when completing tasks, implementing major features, or before merging to verify work meets requirements

Structured external context research — 10-point full checklist (L3-L4), 5-point lite mode (L2), gap discovery protocol for unknowns during implementation.

Credential rotation playbook — consumer inventory, auth-scoped verification, full payload replay, canonical secret paths, rotation log.

Pattern guidance — flag L3+ backlog one-liners that mix a seam/contract concern with an integration/call-site concern; loaded by /dr-prd and /dr-plan.

Authentication, authorization, input validation, data protection, dependency safety. Use for security review or when handling secrets and user data.

Canonical S1–S11 security rule reference cited from AGENTS.md § Security Mandate. Load for plan/qa/compliance/do touching shipped artifacts.

Orchestrator for runtime-aware self-verification (manual /dr-verify). Tri-layer: deterministic shell, peer-review, runtime dispatch.

Consumer contract for /dr-continue — reads session artefact in a clean window, re-verifies every claim, emits banners, and routes to /dr-next or /dr-auto.

Producer contract for /dr-save — writes self-contained session artefacts with append-only semantics, per-layer cap, and claim-provenance enforcement.

Producer contract for per-task stage snapshots — final operator-visible /dr-* response persisted to datarim/snapshots/{TASK-ID}.snapshot.md with overwrite semantics. Single producer touchpoint via cta-format.md.

Reviews an artifact for organisation, completeness, and internal consistency — does it hold together as a document. Loaded at /dr-plan and /dr-qa.

Demands schema-unit + wrapper-path tests when API-side structured-output validation is added on top of an existing post-processing pipeline.

Use when executing implementation plans with independent tasks in the current session

Use when encountering any bug, test failure, or unexpected behavior, before proposing fixes

Technology guidance and stack selection: starting points, viable alternatives, decision-making method, and proposal template for new projects and modules.

Mandatory gate: verify the change on the test env (backend + frontend) autonomously before prod prep or archive. Blocks /dr-qa, /dr-compliance, /dr-archive.

Testing pyramid, frameworks (Vitest, Playwright, bats), mocking rules, Live Smoke-Test Gate for raw SQL. Use when writing or reviewing tests.

Use when starting feature work that needs isolation or before executing plans. Ensures an isolated workspace via native tools or git worktree fallback.

Native shell recipes for common operations. Load this entry first, then only the fragment needed for the specific utility category.

Pattern guidance — when a verification-AC group mixes deterministic axis (rule match) and statistical axis (rate threshold over a window), split upfront into two groups.

Pre-implementation gate proving every runtime-command V-AC (docker exec / curl / kubectl / systemctl / live DB query) can actually PASS before /dr-do.

Use before claiming work complete, fixed, or passing — and before committing or opening PRs. Run verification commands and confirm output first.

Visual index for Datarim maps. Load this entry first, then only the diagram fragment needed for routing, stage flow, or dependency orientation.

Interactive task-spec wizard — a thin orchestrator over discovery and consilium that turns a rough brief into a structured spec plus a graph artefact.

Content creation and go-to-market — research, drafting, editing, publication, SEO, analytics, launch. Loaded by writer and editor agents.

Use when you have a spec or requirements for a multi-step task, before touching code