Samples & Repositories
Inspect the architecture patterns, motivation, and tangible outputs of the open tooling and agent systems we engineer and maintain.
agent-skills
Fundamental skills for agents doing all kinds of work.
Motivation & Why We Built It
Most agentic work includes fundamental tasks that need hardening: safely moving files, writing in a particular tone of voice, and navigating your file system layout. This plugin was built to provide our agents a common set of foundational skills that can be used across all of our agentic workflows.
Problem Solved
Provides a common baseline for file management, writing, and directory navigation. Understand how to manage a PARA vault, but does not force its use.
Architectural Highlights
- Three foundational skills (manage-file-operations, para-vault, write-like-me) sharing a single startup foundation.
- Non-destructive file operations with pre-flight collision checks, relative link repair, and restorable archive snapshots.
- Cross-compatible with Claude Code, Codex, Antigravity, Cursor, Hermes Agent, and OpenClaw.
Sample Output Exhibit: PARA Vault Filing, Prior-Version Archival & MOC Navigation Output
Actual workspace artifacts produced by para-vault and manage-file-operations during inbox filing and protected policy revision (Cases C03 & C17):
# 1. Filed working note (2-areas/finance/escalation-procedure.md)
# Inbox intent fields (filing-hint, context, source) stripped from YAML and preserved in body
---
type: note
created: 2026-09-20
document-maturity: raw
---
# Escalation procedure
Use the incident queue.
## Context and source
Required by audit A-17; owner is Mira; applies to vendor exceptions.
Source: Interview with Mira on 2026-09-19.
# 2. Protected canonical policy revision (3-resources/rules/approval-policy.md)
---
type: note
created: 2026-09-20
document-maturity: established
canonical: true
tags: [policy]
task-state: completed
---
# Approval policy
One reviewer must approve every release.
Canonical designation: 2026-09-01T12:00:00Z.
# 3. Companion archive metadata (4-archives/2026.09.29-approval-policy-prior-version-md-archive-metadata.md)
---
type: note
created: 2026-09-29
document-maturity: raw
canonical: false
archived: 2026-09-29
archived-from: 3-resources/rules/approval-policy.md
archive-reason: superseded
---
# Accepted prior-version evidence
[Unchanged prior policy](2026.09.29-approval-policy-prior-version.md) is historical restoration evidence. Its embedded canonical true flag does not confer current authority. Preserved before the explicitly approved one-reviewer revision and metadata migration.daily-tasks
Markdown-first task management and daily-planning tools.
Motivation & Why We Built It
Most task management tools force users into closed SaaS databases. daily-tasks keeps everything in human-readable Markdown files.
Problem Solved
Bridges unstructured notes and structured action items without vendor lock-in or proprietary task databases.
Architectural Highlights
- No database: projects, milestones, tasks, blockers, and daily plans live in editable Markdown files with UUIDv7 identities.
- Deterministic Node.js CLI handles once-per-day carry-forward rollover, DoD closure verification, and write-ahead checkpoints.
- Offline, fully local execution with separated My work, Delegated work, and Follow-up work views.
Sample Output Exhibit: Persisted Daily Plan & Interactive Check-In Screen (2026.10.27-daily-plan.md)
Actual daily plan Markdown document generated by daily-tasks with carry-forward markers (↺) and UUIDv7 projection refs:
---
id: 01a17f0a-159b-7c80-96fa-310a942e565a
type: daily-plan
title: "Daily plan 2026.10.27"
document-maturity: reviewed
timezone: "America/Chicago"
created_at: "2026-10-27T14:00:00.000Z"
updated_at: "2026-10-27T14:00:00.000Z"
---
# Daily plan — 2026.10.27
Open projects: 1 Open tasks: 2 Late tasks: 0
**Timezone:** America/Chicago
**Updated:** Oct 27, 2026, 9:00:00 AM CDT (America/Chicago)
## Selected for today
### My work
#### Website Launch / Content & CRM Sync
| # | Status | Project | Milestone | Task | Due Date |
| --- | ------ | -------------- | ------------------ | ------------------------------ | ---------- |
| 1 | [>] | Website Launch | Content & CRM Sync | ↺ Enforce library route guards | 2026.10.27 |
| 2 | [ ] | Website Launch | Content & CRM Sync | Update documentation | 2026.10.27 |
### Delegated work
#### Website Launch / Content & CRM Sync
| # | Status | Project | Milestone | Task | Due Date |
| --- | ------ | -------------- | ------------------ | ------------------------------ | ---------- |
| 3 | [ ] | Website Launch | Content & CRM Sync | ↺ Audit CRM outbox replay | 2026.10.29 |
### Follow-up work
No work selected for this section.
Legend: [ ] not started · [>] in progress · [x] completed · [-] cancelled · ↺ carried forward
## Blockers
No active blockers.plan-product
Structured product definition, roadmapping, and dependency-ordered feature planning for agentic software workflows.
Motivation & Why We Built It
Modern AI coding agents fail or introduce severe regressions when given open-ended, ambiguous prompts. We built plan-product to enforce a rigorous specification-first lifecycle: breaking products down into domain concepts, atomic features, typed use cases with ASCII process trees, and dependency-ordered roadmap steps before code is written.
Problem Solved
Eliminates scope creep, agent hallucination on requirements, and unordered delivery dependencies by compiling high-level ideas into verifiable Markdown contracts.
Architectural Highlights
- Strict file-system contract (docs/product.md, docs/concepts/, docs/features/, docs/roadmap.md).
- Predictable Model v2 numbering schema (C-nn for concepts, F-gg.nn for features, UC-gg.nn.uu for use cases, R-pp.ss for roadmap steps).
- Used in Nyssa AI client engagements and internal product builds to accelerate delivery velocity.
Sample Output Exhibit: Use Case Specification Artifact (docs/features/f-12.04/uc-12.04.01-sync-appointment-status-with-crm.md)
Actual Model v2 use case specification and ASCII process tree generated by plan-product for this repository:
---
feature: F-12.04
id: UC-12.04.01
status: planned
---
# UC-12.04.01 Synchronize Appointment Lifecycle with CRM
## Actor
System Automation / Scheduling Webhook Listener / CRM Port (`ICrmPort`).
## Goal
Synchronize calendar appointment lifecycle transitions in real time to the CRM, ensuring deals advance accurately and automated outbound sales sequences are paused upon booking.
## Process
```text
System synchronizes appointment lifecycle event with CRM
+-- Starting context: appointment event emitted or webhook received
+-- Entry route: internal event bus or /api/webhooks/scheduling/[provider]
+-- Trigger: appointment status transition occurs
+-- Step 1: System writes event payload to local write-ahead outbox ledger:
| +-- Normalizes event to canonical AppointmentLifecycleEvent schema
| \-- Assigns sync task status = 'pending'
+-- Step 2: System queries CRM via ICrmPort for contact by attendee email:
| +-- If contact exists as Lead: executes conversion to Contact, Account, and Deal #1
| \-- If contact already exists as Contact: locates existing Deal #1
+-- Step 3: System branches on appointment event type:
| +-- Case 'scheduled':
| | +-- Advances Deal #1 stage to 'session_scheduled'
| | +-- Sets Next_Meeting_Date and Meeting_Link attributes
| | +-- Attaches intake questionnaire responses as structured Deal notes
| | \-- Notifies Email Campaign Engine (F-10.03) to pause active sequences
| +-- Case 'rescheduled':
| | +-- Updates Next_Meeting_Date on Deal #1
| | \-- Appends note with previous and new meeting timestamps
| +-- Case 'cancelled':
| | +-- Sets Deal #1 stage to 'session_cancelled'
| | \-- Creates follow-up task for sales representative
| +-- Case 'completed':
| | +-- Sets Deal #1 stage to 'session_completed'
| | \-- Generates consultant action item to submit session recap
| \-- Case 'no_show':
| +-- Flags Deal record with Attendance_Status = 'NO_SHOW'
| \-- Triggers re-booking automated email sequence or task
+-- Step 4: System updates local outbox record status to 'completed'
\-- Success: CRM Deal synchronized, notes attached, email campaigns paused
```
## Failure Cases
- **CRM Outage / Rate Limit**: CRM API fails during sync; mutation is captured in write-ahead outbox queue and retried with exponential backoff.
- **Unmatched Contact Record**: Booking contains new email without matching CRM record; system converts contact to Lead/Account/Deal automatically before updating stage.
- **Out-of-Order Webhook Delivery**: Late-arriving `scheduled` webhook received after `rescheduled`; timestamp evaluation ignores stale event states.plugin-builder
Multi-host agent plugin authoring, architectural review, and evaluation system.
Motivation & Why We Built It
Building agent plugins that stay clean, token-efficient, and portable across Claude, Codex, Antigravity, Cursor, Hermes, and OpenClaw is error-prone without strict structural guardrails. plugin-builder guides the entire plugin lifecycle from topology selection and scaffolding to bundle assembly, adversarial review, and multi-host evaluation receipts.
Problem Solved
Prevents prompt bloat, trigger collisions, non-canonical file sprawl, and unverified host support claims across multi-host agent plugins.
Architectural Highlights
- Supports three canonical plugin topologies: Public Suite of Skills, Encapsulated Facade, and Autonomous Swarm.
- 6-gate lifecycle: Shape & Scaffold, Authoring, Unit Verification, Host Bundle Assembly (assemble.py & validate.py), Adversarial Review, and Multi-Host Matrix Release Gate.
- 10-dimension architectural audit generating concrete remediation plans in docs/refinement-plans/.
Sample Output Exhibit: Plugin Shape & 10-Dimension Architectural Review Output (docs/refinement-plans/plugin-builder-review.md)
Actual architectural review and lifecycle topology report generated by plugin-builder:
# Plugin Architectural Review & Refinement Plan: `plugin-builder`
> [!IMPORTANT] Scope Directive
> This document records the architectural evaluation and refinement plan for `plugin-builder`. It identifies compliance, risks, and concrete remediation steps across the canonical plugin authoring standards.
**Evaluation Date**: `2026-10-17`
**Evaluator**: `plugin-builder`
**Current Version**: `0.4.0`
**Plugin Archetype**: `Pure Skill`
**Current Architectural Style**: `Encapsulated Facade` (with Autonomous Swarm specialists `adversary` and `eval`)
---
## Lifecycle Tree
```text
Plugin Build Lifecycle
│
├── 1. Plan & Setup ──> workflows/shape.md & workflows/scaffold.md (docs/plugin-shape.md)
│ └── Topology ──> references/architectural-styles.md
├── 2. Authoring ──> workflows/create-skill.md (contracts, SKILL.md, src/)
│ └── Reorganization ──> workflows/reorganize.md (topology migration)
├── 3. Verification Gate ──> test/ (unit & use_case) and evals/suites/waza/ (mandatory triggers)
├── 4. Packaging ──> workflows/package.md (assemble.py & validate.py)
├── 5. Adversarial Review ──> adversary (probes token bloat, RSS memory, isolation)
└── 6. Release Gate ──> eval (matrix verification & evals/LATEST.md)
```
## Dimension-by-Dimension Findings (Excerpt)
### 1. Topology & Architectural Style
- **Status**: `Compliant`
- **Observed Style**: `Encapsulated Facade` (with `Autonomous Swarm` subagents `adversary` and `eval`)
- **Analysis**:
- Global system prompt footprint: Only `skills/plugin-builder/SKILL.md` (81 lines) carries discoverable skill frontmatter. Internal procedures (`workflows/*.md`) have no YAML frontmatter and load on demand.
- Trigger collision & surface clutter: Zero intra-plugin skill collision.
### 4. Package Organization & Manifest Conformance
- **Status**: `Compliant`
- **Manifest Checks**:
- Root `plugin.json` schema conformance and identity: `Pass` (`v0.4.0`, `PolyForm-Small-Business-1.0.0`).
- Release artifact hygiene (zero leaks of `evals/`, `docs/`, `.temp/` into bundles): `Pass`.Want custom skills or agents built for your repo?
We deploy production-tested agents, custom MCP tools, and deterministic pipelines directly into your private codebase during monthly retainer sprints.