j-mauro-r/SAT-Dengue-V2 / .agents/skills/deepworkplan
deepworkplan
VerifiedDeepWorkPlan — turn any repo AI-first and run Deep Work Plans. Routes to create, execute, refine, resume, status, verify, upgrade, and repo-onboarding sub-skills based on intent. Use when the developer wants to plan, execute, manage, or verify structured multi-task work, or make a repository AI-agent-ready.
spec/V6_RESOURCES.md
View on GitHub# V6 Resources — Envelope Reserves, Exhaustion and Host Capability Negotiation
> **Status: v6 line.** This document defines the v6 resource contract: the
> **host capability negotiation**, the **counter-source split**
> (journal-observable vs host-only), the **reserve** representation, the
> **exhaustion** grammar and dispatch hold, the **cancellation
> settlement**, and the **routing posture**. The design source is
> `docs/evaluations/v6/ARCHITECTURE_RFC.md` §8 (draft-4); the shipped
> implementation is `shared/resources.py` (Python 3.9+ stdlib, composing
> the envelope accounting of `shared/scheduler.py` and the record
> discipline of `shared/ledger.py` — it never duplicates either). The v5
> standard is untouched: these rules bind **v6 new plans only**.
All documents in this spec use RFC-2119 language. A limit the host cannot
meter is **advisory, not enforced**: enforcement parity across hosts is
never claimed.
## 1. Host capability negotiation
* The runtime host states its **abilities** as a closed set:
`stop_agent`, `meter_spend`, `meter_tokens`, `meter_wall_clock`,
`cancel_children`, `model_routing`, `subagents`, `telemetry`. Abilities
are distinct from the contract's permission **grants** (§3.2 of the
contract spec): a grant is authority, an ability is what the host can
actually do.
* An unknown ability key MUST be refused — a capability is never invented.
* An unstated ability is `false`. The **minimal host** (no model
switching, no subagents, no telemetry, no stop) is a supported degraded
posture, never an error; every limit that needs the missing ability
degrades to advisory with the missing ability named.
* From the 7.0.0 line, enabled and detected addons MAY contribute
abilities at runtime on top of this declaration — never persisted, never
invented — per [`V7_ABILITIES.md`](V7_ABILITIES.md); this section is
otherwise unchanged.
## 2. Counter sources
* **Journal-observable counters** (wall-clock span, gate retries,
dispatch calls) are computed from the plan's own records; no host
ability is required to enforce them.
* **Host-only counters** (spend, provider tokens) require the matching
`meter_*` ability and an **observed** `resource_sample` from a host
adapter (A1: only a host_adapter actor reading a real meter mints
observed samples). A metered host with no sample on record enforces
the **pending side only** — a missing counter is not free usage.
* A unit outside every known family MUST degrade to advisory with the
unit named; a meter is never invented.
## 3. Reserves
* A limit MAY declare `reserve` (a number, `0 <= reserve <= limit`). The
dispatchable ceiling is `limit - reserve`; a refusal fires when
`spent + pending > limit - reserve`. Reserves exist so verification,
retry and resume cannot be starved by ordinary work.
* The ceiling comparison is **runtime semantics** (it depends on the
sibling `limit` value, which JSON Schema cannot express); both halves
still reject a malformed (non-numeric, negative) reserve.
## 4. Exhaustion
* Hitting a limit MUST persist an **incomplete checkpoint** — the journal
and snapshot already are that checkpoint — and stop dispatch. The stop
is recorded as an observation with the grammar
`LIMIT: exhausted <limit-id>: <detail>`; only a declared limit may be
exhausted (an undeclared counter is never recorded).
* An exhaustion observation **can never satisfy an acceptance criterion**
(the ledger counts only `gate_run` events); exhaustion structurally
cannot mint completed evidence.
* Dispatch stays held until an explicit `LIMIT: recovered <id>` record;
recovery is never silent.
* Advisory limits surface, never hold.
## 5. Cancellation settlement
* An in-flight reservation (an authorized adaptation's declared impact)
settles **exactly once**: `RESERVATION: <key>: released` (the cancelled
child never consumed the budget) or `RESERVATION: <key>: committed`.
* Identical replay is idempotent (content-keyed dedup). A second
settlement with a **different** disposition is the double-charge
ambiguity and MUST be refused, never arbitrated.
* Pending accounting subtracts released reservations; with zero
settlements it equals the envelope core's pending exactly (drift-guarded
by tests).
## 6. Routing posture
* The default posture is **fixed-model**. Model-tier switching requires
BOTH the contract grant `model_routing` AND the host ability of the
same name; each refusal names the missing side(s).
* Parallel dispatch requires the `agent_delegation` grant AND the host
`subagents` ability; otherwise the posture is sequential-only.
* A routing posture is a permission statement, **never an efficacy
claim**: switching being authorized says nothing about whether
switching helps.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis. | Stars: 472k | Updated | Claude CodeCodex | Verified | Install |
| diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness. | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependencies | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is available | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |