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/CONFIG.md
View on GitHub# CONFIG.md — DeepWorkPlan configuration file (Normative)
> Status: **7.0.0**. Defines `.dwp/config.json` and
> `~/.dwp/config.json`: one file shape, one parser
> ([`../shared/config.py`](../shared/config.py)), two keys — `benchmark`
> ([`BENCHMARK.md`](BENCHMARK.md) §1) and `addons` (the addon registry,
> §3 below). Published as
> [`schema/dwp-config-v1.schema.json`](schema/dwp-config-v1.schema.json)
> (`https://deepworkplan.com/schema/dwp-config/v1.json`). RFC-2119 language.
**Governing principle.** The methodology works with no configuration file
and no addon. Nothing in this file can make a plan, a flow or conformance
depend on an addon: the registry records what a repository *opted into*, so
flows may **offer** or **amplify** — never require.
## 1. Locations and precedence
| Precedence | Path | Scope |
|---|---|---|
| 1 (highest) | `<repo-root>/.dwp/config.json` | the repository that owns the plan (the plan's `.dwp` ancestor), or the repository a flow runs in when there is no plan |
| 2 | `~/.dwp/config.json` | every repository of the local user |
| 3 | absent | everything disabled |
**The repository file is tracked** so teammates and CI read the same
registry (F-04): onboarding writes the `.gitignore` rule `.dwp/*` +
`!.dwp/config.json` — plans, onboarding notes and benchmark records stay
ignored. A repository with a plain `.dwp/` rule stays conformant; its
registry is then local. The file holds decisions, never secrets.
Both files share one shape:
```json
{
"benchmark": { "enabled": true, "learnings": true },
"addons": {
"ai-diff-reviewer": { "enabled": true, "version": "v3.3.14" },
"herdr": { "enabled": true, "version": "v1.0.0" },
"agentkit": { "enabled": false },
"devcontainer": { "enabled": true, "version": "v1.0.1" },
"vim": { "enabled": true, "version": "v1.0.0" },
"dailybot": { "enabled": true, "version": "v3.23.17" }
}
}
```
- An optional top-level `"schema"` MAY name
`https://deepworkplan.com/schema/dwp-config/v1.json`; readers do not
require it. Unknown top-level keys are ignored (forward compatibility).
- **Resolution differs by key.** `benchmark` resolves **wholesale**: a
repository file that carries the object overrides the user file for both
of its flags (`BENCHMARK.md` §1). `addons` resolves **per addon key**:
the repository entry for a key wins over the user entry for that key; a
repository file that omits a key defers to the user file for it.
## 2. Fail-closed reading (both keys)
- A missing file is the ordinary "omits everything" case — no warning.
- An unreadable path or invalid JSON: the file contributes nothing, with
exactly **one** warning naming the file and the reason; the other file
still applies.
- A wrong-typed value disables only what it affects, with **one** warning
naming the file, the key and the reason (`benchmark` keeps its §1 rules;
`addons` per §3).
- A reader MUST NOT raise, MUST NOT abort a flow, and MUST NOT be silent
about a value it discarded.
## 3. The addon registry (`addons`)
- **Key set.** The registry keys are the **in-pack addon directory names**
(`addons/<key>/`; ecosystem contract amendment A1) — today
`ai-diff-reviewer`, `agentkit`, `dailybot`, `dependency-upgrade`,
`design-system`, `devcontainer`, `herdr`, `vim`. Each addon ships an
`addon.json` descriptor whose `key` equals its directory name
([`ADDONS.md`](ADDONS.md) §7). The reader derives the set by listing
directories; it never opens a file inside an addon to do so. A key
outside the set is **ignored with one warning** (a newer pack's addon, a
typo) — never an error.
- **Entry shape (closed).** `enabled` (boolean, **required to enable**;
only `true` enables), optional `version` (string matching
`^v\d+\.\d+\.\d+(-[0-9A-Za-z.]+)?$`) and optional `note` (one line,
1–200 characters — e.g. "accepted; machine-level install deferred").
An enabled addon that is not detected and carries a note is reported as
**deferred** (`resources.py abilities` → `deferred`), without a warning
(F-05). Any other field, a missing
`enabled`, a non-boolean `enabled` or a malformed `version` makes that
entry **not enabled** with one warning naming the file, key and reason.
A malformed entry in the repository file decides its key (fail-closed);
it does not fall through to the user file.
- **`"addons"` itself** must be an object; anything else contributes no
entry from that file, with one warning.
- **Absent = not enabled.** Nothing is enabled by default. Only what
onboarding recorded with consent, an addon's own install step, or the
human enables anything.
- **The AI Diff Reviewer baseline is unchanged.** Its local review stays
part of the Final Review (`ADDONS.md` §6.5); a missing installation is the
recorded `local reviewer not installed` finding whether or not the
registry names it. The registry adds no requirement.
- **`version` is informative in 7.0.0.** `verify` MAY warn when it differs
from the installed product's observed version; it MUST NOT fail on it.
- **Enabled is necessary, not sufficient.** An addon contributes abilities
only when it is enabled **and** its descriptor's detection succeeds with
a compatible interface major (`V7_ABILITIES.md`); being listed never
makes a product present.
## 3a. The host capability record (`host`)
The abilities of the host that runs plans here, machine-readable (F-17):
`"host": {"subagents": true, "cancel_children": true}` — the closed v6 set
(`stop_agent`, `meter_spend`, `meter_tokens`, `meter_wall_clock`,
`cancel_children`, `model_routing`, `subagents`, `telemetry`), booleans
only. Per capability the user file wins over the repository file (the
record describes a machine; the tracked file is the team baseline), and an
explicit `resources.py --caps` declaration wins over both; an unstated
capability stays at the all-False floor. An unknown or non-boolean
capability is ignored with one warning — never invented. `resources.py
abilities` reads it and names where each declared capability came from
(`host_declared`). Writer: `config.py host <capability> true|false --repo
<repo>`, run by onboarding when it records the host declaration
([`../onboard/v6.md`](../onboard/v6.md) §1); the prose block in
`AGENTS.md` stays the human summary. `telemetry: true` still needs the
developer's consent; the record states the host can, not that it may.
## 4. Writers and readers
| Role | Who | How |
|---|---|---|
| Writer | `onboard` Phase 7b, on the developer's acceptance of one addon's offer | `python3 <pack>/shared/config.py enable <key> [--version <tag>] --repo <repo>` (a decline writes nothing; an explicit "turn it off" is `disable <key>`) |
| Writer | an addon's own install step, or the human | the same command, or a hand edit |
| Writer | the developer, or an agent they asked, turning field metrics on | `python3 <pack>/shared/config.py benchmark enabled\|learnings true\|false --repo <repo>` (7.1.5): sets one `benchmark` key, preserves the other and every other top-level key, refuses an unknown key or a file it cannot parse. `learnings` still rides on `enabled` (BENCHMARK.md §1). `enable`/`disable` stay addon-only |
| Writer | `upgrade`, after the developer accepts the upgrade | `config.py backfill --repo <repo>` (dry run), then `--write`: every in-pack addon the repository file does not name and that is installed **in the repository** (a repo-relative detect path) is recorded `enabled` with its observed version and a back-fill note (F-15); a machine-level install is reported, never recorded — onboarding offers it. A recorded decision — enabled or disabled — is never changed |
| Reader | `execute`, `create`, `verify`, `status` | `python3 <pack>/shared/config.py show` / `enabled` (`--plan <dir>` resolves the plan's repository) |
| Reader | `shared/benchmark.py` | the `benchmark` key only, at its emission point |
The writer **reconciles**: it preserves every other top-level key and every
other addon entry, writes atomically, and **refuses** (exit 2, nothing
written) an unknown key, a malformed version, or an existing file it cannot
parse — it never overwrites what it cannot read.
The `show` view is `{"benchmark": {...}, "addons": {key: {"enabled",
"version", "source"}}}` with warnings on stderr; `enabled` prints the
enabled keys one per line. Both are read-only.
## 5. Conformance
- A repository with no configuration file is fully conformant, and every
flow behaves identically with no file, with every addon disabled, or with
only unknown keys (`tests/standalone-methodology.bats`).
- The configuration file is never a plan artifact: plans never persist the
registry or the abilities it contributes (`V7_ABILITIES.md`); a plan
records only what it **used**, in its journal.
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 |