// autonomous software organization / version-controlled workforce & policy

HowlFutureWorks

A human-governed autonomous software organization whose roles, workforce, policies, knowledge, evidence and operating procedures are represented as version-controlled desired state.

ORGANIZATION AXIOM // MEMORY IS NOT AUTHORITY. Git defines desired organizational state. Live AI workers are deployed state. Durable knowledge is curated, evidence-backed and recoverable.

This complements the ecosystem-wide principle: INTENT IS NOT AUTHORITY.

[ VIEW ON GITHUB ] [ READ THE CHARTER ] [ HOWL ECOSYSTEM ]

SECTION // 01 Overview & Systems of Truth

HFW-MOD-OVERVIEW

HowlFutureWorks

Defines organizational desired state: positions, AI workforce identities, staffing lifecycle, organizational policy, institutional knowledge, authority boundaries, organizational runbooks, schemas, continuity, and deployed-workforce reconciliation.

HowlPlane

Coordinates execution: tasks, provider routing, delegation, workflow state, execution, recovery/resume, and evidence flow.

HowlFutureWorks governs the organization. HowlPlane governs execution of work. It is not another HowlPlane.

The organization's systems of truth, as declared in organization.yaml and CHARTER.md:

Desired Organization

Git. desired_state_source: git — this repository is the canonical, version-controlled definition of positions, policy and roster.

Deployed Workforce

Live persistent AI members. Grok Bot instances reconciled against workforce/roster.yaml; bot memory is never the source of truth.

Work State

HowlBoard. Listed in organization.yaml's work_state_sources as a source of work/execution state.

[ LIVE SITE ] [ GITHUB ]

Execution State

HowlPlane. Also listed in organization.yaml's work_state_sources; coordinates task routing and execution per CHARTER.md §9.

[ LIVE SITE ] [ GITHUB ]

Policy / Authority

HowlFrame. policy_enforcement: HowlFrame in organization.yaml.

[ LIVE SITE ] [ GITHUB ]

Change Control

HowlChangeOps. Referenced in CHARTER.md §9 as the ecosystem's release/change-control component; no direct runtime integration is implemented in this repository.

[ LIVE SITE ] [ GITHUB ]

Verification

Assurance (internal position) / HowlProof / CI. This repository's own CI (.github/workflows/validate.yml) runs orgctl.py validate and the test suite on every push.

[ LIVE SITE ] [ GITHUB ]

Institutional Memory

workforce/ + knowledge/ + adr/ + runbooks/ — curated, evidence-backed and recoverable, not raw chat transcripts.

SECTION // 02 Organization

HFW-MOD-ORGANIZATION

ROLE ≠ POSITION ≠ EMPLOYEE INSTANCE

Role

Durable responsibility.

Position

Persistent organizational seat and configuration.

Employee

Replaceable worker instance occupying a position.

[ORG // REPORTING TOPOLOGY] HFW_TOPOLOGY

                         HUMAN OWNER
                              │
              ┌───────────────┴───────────────┐
              │                               │
              ▼                               ▼
ENGINEERING MANAGER / ORGANIZER      INDEPENDENT AUDITOR
              │
      ┌───────┼────────┬──────────┐
      ▼       ▼        ▼          ▼
   PRODUCT   R&D    DEV LEAD   ASSURANCE

The Independent Auditor reports directly to the Owner (bots/manifest.yaml: staffing_authority: owner, the same protected status as the Engineering Manager) — not to the Engineering Manager. This preserves audit independence: per CHARTER.md §8, "Audit should be read-only wherever practical and should report directly to the Owner rather than to the team being audited."

SECTION // 03 Founding Workforce

HFW-MOD-WORKFORCE

Six founding persistent positions, per bots/manifest.yaml and CHARTER.md's "persistent roster strategy":

Engineering Manager / Organizer REPORTS TO OWNER

Coordinates organizational reconciliation, staffing, delegation, blockers and health.

Product REPORTS TO ENGINEERING MANAGER

Owns WHAT/WHY, problem evidence, priority and acceptance criteria.

R&D REPORTS TO ENGINEERING MANAGER

Reduces uncertainty through controlled investigation and experiments.

Dev Lead REPORTS TO ENGINEERING MANAGER

Coordinates implementation and appropriate execution-provider routing.

Assurance REPORTS TO ENGINEERING MANAGER

Coordinates logically separate QA and Security verification.

Independent Auditor REPORTS TO OWNER

Independently reconstructs evidence, decisions and process compliance and reports to the Owner.

Execution models such as Claude, Codex, or other providers are not automatically persistent organization members. Positions are staffed by workforce-managed employee instances (workforce/roster.yaml) — a provider is a replaceable executor of a task, not an organizational identity.

SECTION // 04 Governance & Risk

HFW-MOD-GOVERNANCE

Risk tiers, per policies/risk-tiers.yaml and RISK.md. Risk is attached to the action, not the Bot title.

R0 // Observe AUTO

Read-only research, inventory, status, reporting.

R1 // Sandbox AUTO

Isolated worktree changes, disposable experiments, generated files.

R2 // Development AUTO

Branches, pull requests, tests, development CI, issue updates.

R3 // Shared / Release APPROVAL REQUIRED

Merge, staging/shared mutation, publication, external communications, release promotion.

R4 // Critical APPROVAL REQUIRED

Production, destructive bulk action, privilege change, secret operations, irreversible data, security-control change, financial or legal commitment.

Invariants:

NO SELF-ESCALATION

"No agent may self-escalate privileges" (AGENTS.md).

ROLE NAME ≠ SECURITY PRINCIPAL

"Role names are not security principals" (AGENTS.md); enforcement comes from HowlFrame, scoped accounts and repository permissions, not titles.

HIGH-RISK ACTIONS FAIL CLOSED

"High-risk actions fail closed when required policy/approval/audit is unavailable" (AGENTS.md); policies/approvals.yaml: fail_closed_if_required_approval_unavailable: true.

APPROVAL BINDS TO THE ACTUAL ACTION

"High-impact approvals bind to the exact action/target/parameters and expire" (AGENTS.md); policies/approvals.yaml requires a new approval when the action materially changes.

SECRETS DO NOT BELONG IN BOT MEMORY

policies/memory.yaml: secrets_in_memory: forbidden. "Never write secrets into prompts, Markdown, YAML, reports, evidence, or Git history" (AGENTS.md).

HUMAN OWNER RETAINS CRITICAL AUTHORITY

"Human Owner retains final strategic and critical authority" (AGENTS.md).

[ CHARTER.md ] [ RISK.md ] [ SECURITY.md ] [ policies/ ]

SECTION // 05 Knowledge Continuity

HFW-MOD-KNOWLEDGE

[HIERARCHY // PROMOTION LADDER] HFW_MEMORY

conversation
     ↓
task / context checkpoint
     ↓
employee continuity
     ↓
position knowledge
     ↓
company knowledge
     ↓
ADR / runbook / policy

WRITE ONCE, RETRIEVE MANY

Conversations are disposable.

Durable lessons are checkpointed before context loss.

Evidence is referenced instead of copied.

Knowledge is promoted upward.

Lower-level duplicates are compacted.

Raw chats are not institutional memory.

Private reasoning / chain-of-thought is not persisted.

Current numeric budgets, per policies/budgets.yaml and organization.yaml (the website is descriptive, not authoritative — read those files for current values):

Budget	Value	Source field
Context checkpoint threshold	75%	`policies/budgets.yaml`: `defaults.context_checkpoint_threshold_percent`
Checkpoint maximum	4,000 chars	`policies/budgets.yaml`: `knowledge_artifacts.checkpoint_chars`
Handoff maximum	6,000 chars	`policies/budgets.yaml`: `knowledge_artifacts.handoff_chars`
Position knowledge maximum	8,000 chars	`policies/budgets.yaml`: `knowledge_artifacts.position_knowledge_file_chars`
Employee continuity maximum	24,000 chars	`organization.yaml`: `employee_continuity_budget_chars`

SECTION // 06 Bootstrap

HFW-MOD-BOOTSTRAP

[FLOW // GIT TO VALIDATED ORGANIZATION] HFW_BOOTSTRAP

GIT DESIRED STATE
       ↓
ENGINEERING MANAGER / ORGANIZER
       ↓
LIVE WORKFORCE RECONCILIATION
       ↓
PRODUCT / R&D / DEV LEAD / ASSURANCE / AUDITOR
       ↓
VALIDATED AUTONOMOUS ORGANIZATION

[BASH // ORGCTL_INSPECTION]

python tools/orgctl.py about
python tools/orgctl.py validate
python tools/orgctl.py org-status
python tools/orgctl.py list-bots
python tools/orgctl.py list-workforce
python tools/orgctl.py render-all

[BASH // KNOWLEDGE_CHECKPOINT_EXAMPLE]

python tools/orgctl.py checkpoint bot-devlead-0001 \
  --reason context_budget \
  --stable "Preserve the durable result, not the conversation." \
  --source-ref "adr/example.md"

No secrets are ever written to a checkpoint — orgctl.py checkpoint always records contains_raw_chain_of_thought: false and rejects snapshots that exceed the checkpoint budget.