186 lines
11 KiB
Markdown
186 lines
11 KiB
Markdown
# SCOPE
|
|
|
|
> Reviewed 2026-08-20 against the implemented Helix Forge loop and the
|
|
> cross-repository leadership/workforce architecture discussion. For the
|
|
> stable purpose and desired organizational outcome, see `INTENT.md`.
|
|
|
|
## One-liner
|
|
|
|
Agent-workforce performance and lifecycle management: learn from agent
|
|
execution, improve the fleet, and—once the proposed contracts are
|
|
implemented—plan and govern which agents fill which roles within
|
|
leadership-owned goals and delegated resource envelopes.
|
|
|
|
## Core Idea
|
|
|
|
`agentic-resources` treats AI agents as a managed digital workforce rather
|
|
than as isolated prompts or interchangeable runtime processes.
|
|
|
|
Its implemented Helix Forge subsystem captures coding sessions, finds
|
|
repeated friction and successful techniques, curates and distributes reviewed
|
|
solution patterns, measures their effectiveness, and produces fleet
|
|
retrospectives. This is already the performance-and-development loop described
|
|
by `INTENT.md`.
|
|
|
|
The recommended next boundary extends that evidence base into agent inventory,
|
|
selection, assignment, capacity planning, development, portfolio review,
|
|
rebalancing, and retirement. This repository may translate authorized goals
|
|
and constraints into a workforce plan, but it does not create organizational
|
|
strategy, leadership authority, money, technical capacity, or execution
|
|
permissions.
|
|
|
|
## Current Implemented Scope
|
|
|
|
- Capture and normalize Claude, Codex, and Grok coding sessions.
|
|
- Retain session digests and raw evidence within explicit storage budgets.
|
|
- Detect recurring failure, friction, cost, and success signals across the
|
|
fleet.
|
|
- Curate evidence-backed solution patterns through a review and approval flow.
|
|
- Render approved patterns as agent-flavor-specific, reviewable distribution
|
|
proposals.
|
|
- Measure before/after pattern effectiveness and fleet trends.
|
|
- Publish weekly retrospectives and correlate fleet-level Helix Forge evidence
|
|
with project-level KaizenAgentic metrics by reference.
|
|
|
|
## Recommended Workforce-Management Boundary
|
|
|
|
The following concerns fit the intent and existing performance subsystem, but
|
|
their contracts and implementation do not yet exist in this repository:
|
|
|
|
- Maintain an inventory of available agent capabilities, instances, status,
|
|
lifecycle stage, and evidence of fitness.
|
|
- Own the `RoleAssignment` control contract: which agent instance fills which
|
|
part of a role for which goals, under which leadership mandate and resource
|
|
envelope.
|
|
- Express execution needs through a versioned `glas-harness` profile reference
|
|
or compatible requirements contract, rather than coupling assignments to one
|
|
concrete rein or base model.
|
|
- Match agents, role duties, and Kaizen blueprints using explicit suitability
|
|
evidence; record unstaffed or deliberately human-owned duties.
|
|
- Allocate concurrent agent and duty capacity within authoritative financial,
|
|
technical-resource, token, time, and human-attention constraints.
|
|
- Aggregate outcomes, costs, refusals, and escalation quality by goal, role,
|
|
assignment, and duty.
|
|
- Maintain performance and development plans and recommend continue, coach,
|
|
refine, reallocate, pause, or retire decisions.
|
|
- Run portfolio reviews and propose rebalancing while requiring the authorized
|
|
leadership source to approve goal reprioritization, material envelope
|
|
changes, or wider authority.
|
|
|
|
Consumer-owned assignment/instance manifests should remain the durable source
|
|
initially; this repository can validate, index, plan, and review them without
|
|
prematurely requiring a central workforce service.
|
|
|
|
## Leadership and Workforce Relationship
|
|
|
|
Leadership and workforce management are adjacent control layers, not synonyms:
|
|
|
|
```text
|
|
organization intent
|
|
-> leadership-owned StrategicGoal + LeadershipMandate
|
|
-> delegated ResourceEnvelope
|
|
-> agentic-resources workforce plan + RoleAssignment
|
|
-> Kaizen blueprint + consumer-owned AgentInstance
|
|
-> activity-core scheduled work
|
|
-> glas-harness profile resolution
|
|
-> selected rein + base-model + sandbox constellation
|
|
-> outcome, cost, evidence, refusal, escalation
|
|
-> leadership review + workforce rebalance + blueprint improvement
|
|
```
|
|
|
|
- A `StrategicGoal` states the desired outcome, priority, horizon, and success
|
|
measures. Its authoritative source is the accountable organization or
|
|
engagement control plane.
|
|
- A `LeadershipMandate` states who may decompose goals, direct work, make which
|
|
decisions, delegate, pause, and escalate. A role title or agent prompt never
|
|
creates this authority.
|
|
- A `ResourceEnvelope` composes provenance-bearing limits from leadership,
|
|
`fin-hub`, `resource-control`, and runtime policy. Unknown capacity or budget
|
|
is never treated as unlimited.
|
|
- A `RoleAssignment` is the workforce decision within those inputs. It binds a
|
|
specific agent instance to selected role duties and records evidence,
|
|
limits, lifecycle, and review state.
|
|
|
|
Agentic-resources therefore advises and administers the agent workforce;
|
|
leadership remains accountable for direction, priority, material commitments,
|
|
and expansion of authority.
|
|
|
|
## Ecosystem Boundaries
|
|
|
|
| Concern | Owning source | Relationship to agentic-resources |
|
|
|---|---|---|
|
|
| Organizational intent, goals, priorities, and leadership mandates | Contextual organization/engagement control source; for example `binky-control` for Binky Hedgehog GmbH | Consume stable references; return portfolio evidence and recommendations; never invent or silently reprioritize strategy |
|
|
| Organizational role definitions | `role-engine` | Consume pinned role revisions and duty/authority boundaries; do not redefine roles or grant permissions |
|
|
| Agent blueprints, protocols, and project metrics | `kaizen-agentic` | Consume versioned blueprints and run evidence; supply assignment and performance context |
|
|
| Agent workforce inventory, assignment, capacity, development, and portfolio review | `agentic-resources` (recommended boundary; not yet implemented beyond performance learning) | Own the workforce plan and assignment lifecycle |
|
|
| Human workforce lifecycle and mixed-team concerns | `human-resources` | Share compatible vocabulary and handoff contracts; keep human-sensitive, employment, welfare, and compensation records separate |
|
|
| Monetary budgets, commitments, burn, runway, and AI-plan booked cost | `fin-hub` | Consume authoritative constraint signals and publish attribution; do not copy the financial ledger or approve spending |
|
|
| Resource identity, demand, capacity, utilization, allocation evidence, and technical economics | `resource-control` | Consume planning/evidence signals; do not procure, operate, or invent capacity |
|
|
| Scheduling and durable claimable work | `activity-core` | Supply active assignment/duty context; do not own the clock, queue, retry, or task lifecycle |
|
|
| Harness contract, execution profiles, backend routing, sandbox coordination, and gateway evidence | `glas-harness` | Supply an approved, versioned profile or execution requirements; let Glas resolve the concrete rein and sandbox constellation; consume its unified result evidence |
|
|
| Concrete inner agentic loop and backend-specific credentials/policy | Selected rein under `glas-harness`; currently including `rein-aharness` and `rein-openweights` | Treat a rein as a replaceable execution backend, not the workforce architecture boundary; `rein-aharness` is the live Claude-Code-CLI rein, not the meta-framework |
|
|
| Identity, credentials, and authorization | Identity, OpenBao/credential-routing, and policy systems | Reference routes and policy profiles only; never store secrets or treat organizational wording as technical access |
|
|
| Base-model/provider execution and inference routing | `llm-connect`, the selected rein, and provider adapters under their declared boundaries | Select through the approved Glas execution profile/route; consume usage/outcome evidence; do not hard-code a vendor model into workforce records or become a model gateway |
|
|
| Durable coordination read model | Custodian State Hub | Publish/index work and evidence; do not replace source files or hub ownership |
|
|
|
|
## Out of Scope
|
|
|
|
- Setting organizational strategy, goals, or cross-goal priorities.
|
|
- Granting leadership mandates, technical authorization, credentials, or tool
|
|
permissions.
|
|
- Owning human personnel records, hiring, compensation, employment decisions,
|
|
welfare, or human performance surveillance.
|
|
- Owning the canonical role catalog or treating a role definition as a live
|
|
assignment.
|
|
- Owning monetary budget truth, payment/commitment approval, procurement,
|
|
infrastructure inventory, or provider capacity.
|
|
- Owning schedules, queues, workflow/task lifecycle, LLM sessions, or runtime
|
|
policy enforcement.
|
|
- Selecting a concrete rein or base model outside the approved
|
|
`glas-harness` profile/route, or treating `rein-aharness` as the universal
|
|
execution architecture.
|
|
- Automatically applying self-modifications or retiring live agents without
|
|
the review and authority required by the relevant mandate.
|
|
- Replacing KaizenAgentic, role-engine, activity-core, `glas-harness`, its
|
|
concrete reins, State Hub, fin-hub, or resource-control with a monolithic
|
|
agent-management service.
|
|
|
|
## Intent Relationship
|
|
|
|
`INTENT.md` describes the desired organizational outcome: a trustworthy,
|
|
antifragile agent ecosystem that improves performance and augments human
|
|
ingenuity through lifecycle management, oversight, and continuous alignment.
|
|
It is deliberately broader than this repository's implementation ownership.
|
|
|
|
This scope realizes that intent through a federated operating model. The repo
|
|
owns the agent-workforce performance and lifecycle concern while consuming
|
|
strategy, role, budget, resource, policy, runtime, and human-workforce facts
|
|
from their authoritative systems.
|
|
|
|
## Current State
|
|
|
|
- **Implemented:** the Helix Forge capture -> detect -> curate -> distribute ->
|
|
measure loop, weekly retrospectives, and Kaizen metrics correlation.
|
|
- **Completed work:** `AGENTIC-WP-0001` through `AGENTIC-WP-0011`; there are no
|
|
active normal workplans at the time of this review.
|
|
- **Architecture proposal:** accept agentic-resources as the workforce and
|
|
assignment owner, subject to cross-repository ratification.
|
|
- **Execution-layer implementation:** `GLAS-WP-0004` now provides strict
|
|
versioned requests/profiles, deterministic rein-registry validation,
|
|
explicit CLI/gateway selection, and one normalized evidence envelope.
|
|
`rein-aharness` is one current backend; workforce contracts preserve freedom
|
|
to choose a different rein, sandbox, tool policy, and base-model
|
|
constellation by changing the approved Glas profile reference. The
|
|
Claude-backed and open-weight profiles both have live success, real-commit,
|
|
and sandbox-cleanup evidence through the same Glas contract.
|
|
- **Implementation gap:** agent inventory, `RoleAssignment`, capacity
|
|
allocation, goal/resource-envelope correlation, development plans, and
|
|
portfolio/retirement workflows are not implemented yet.
|
|
|
|
## Getting Oriented
|
|
|
|
1. `INTENT.md` — stable organizational purpose.
|
|
2. `docs/PRD-helix-forge.md` — implemented product boundary and architecture.
|
|
3. `session_memory/README.md` — current operational capabilities.
|
|
4. `docs/DESIGN-session-memory.md` — detailed design and Kaizen correlation.
|
|
5. `workplans/` — implementation history and future work.
|