The Agentic Operations Package
A portable operating system for traceable AI data work — canonical core, Obsidian vault, auditable workflows, skills, harness and adapters.
A fluent answer is not yet reliable operation. The package separates knowledge, permission, execution, review and human approval — local-first, exportable, and not bound to a single model or platform brand.
One canonical core, many adapters
Five separated levels
| Level | Content | Role | Boundary / gate |
|---|---|---|---|
| Knowledge | canonical core, sources, decisions | planner · researcher | a source carries authority, not permission |
| Permission | task contract, capability grant, data zone | — | tool access does not broaden scope |
| Execution | workflows, skills, file change | executor | only a planned, verifiable step |
| Review | 27 evaluation contracts, review, provenance | reviewer | a golden run is not a pilot |
| Approval | approval gate | gatekeeper | external/irreversible effect only after a human gate |
Six portable platform kits
Task contract
task_id: "TASK-YYYYMMDD-001"
title: ""
owner: "Furkan Sakızlı"
outcome: ""
operating_mode: "explore | design | produce | review | approval-preparation"
data_zone: "open | internal | sensitive | highly-sensitive"
sources_required: []
sources_excluded: []
scope_in: []
scope_out: []
protected_paths: []
permitted_actions: []
prohibited_actions: []
tool_allowlist: []
capability_grants: []
eval_suite_ref: ""
required_roles: []
external_action: "denied | preview-only | allow-once-after-gate"
acceptance_criteria: []
verification: []
approval_gate_id: ""
handoff_path: ""
# The contract applies only to the named task. Tool access is not
# permission; scope, zone or external effect must not grow implicitly.What explicitly does not happen
- no autonomous external effect, publication or deployment without a complete human gate
- no secrets, production credentials or personal data inside the package
- no silent equating of plan id and export number
- no loading of all articles into every agent context
- no adapters before the canonical core and harness are complete
- no storing of raw chain-of-thought — only auditable evidence, decisions and tests
Core principle
A new operator recognises without the chat history: the task, state, sources, decisions, rights, test path, open gates and the next step. Tool access does not broaden a task; external or irreversible effect stays behind a visible human gate.
The building blocks in detail
Canonical core
Ten versioned Markdown contracts keep the task, sources, decisions, change and approval true in one place.
- Project Brief, Source Index, Decisions, Acceptance
- Task Contract, Change Protocol, Approval Gate
- Context Manifest, Exit & Recovery, Handoff
every task is reconstructable without chat
Operational Obsidian vault
A navigable knowledge and article structure as the editorial source for all further packages.
- HOME, MOCs and registers
- 37 article metadata notes
- frontmatter, data cards and templates
knowledge is linked, classified and exportable
Nine workflows
The operational data and change lifecycle with triggers, states, human gates, stop signals and recovery.
- Source Intake, Harmonisation, Knowledge Promotion
- RAG Ingestion, Gap Analysis, Agentic File Change
- Review & Release, Incident & Recovery, Archive & Exit
each flow has an explicit state model
Nine portable skills
Platform-neutral procedure packs — part of the download, but not automatically installed in an agent environment.
- Source Verification, Metadata Normalisation, Provenance Trace
- Chunking Audit, Retrieval Evaluation, Gap Detection
- Safe File Change, Approval Preparation, Handoff Generation
each skill is structurally validated
Harness & evaluations
Roles, an event loop and control surfaces separate planning, execution, review and decision.
- five roles and an auditable event loop
- capability registry, five data zones, default-deny tool allowlist
- 27 executable eval contracts with fixtures and metrics
the golden run passes reproducibly
Four adapter templates
Short, native entry points that reference the canonical core and never broaden its rights.
- Codex (AGENTS.md) and Claude Code (CLAUDE.md)
- Cursor (agentic-project.mdc) and Copilot
- semantic checksum, installation tests, discovery & rollback
exactly the active adapter is installed
Six platform kits
Portable templates for IDEs, messaging agents, browser projects, research and harnesses — not installed integrations.
- IDE, OpenClaw and Hermes Operations
- Browser Projects and NotebookLM Research
- shared invariants and start boundaries
every material session ends with a handoff
Download
Agentic Operations Package — core, vault, workflows, skills, harness & kitsZIP · 172 KB · Members onlyBecome a member to download →The full package is reserved for members. The text on this page stays free; the download (canonical core, Obsidian vault, workflows, skills, harness and adapters) opens only for signed-in members.
0 comments
● Loading comments…