j-mauro-r/SAT-Dengue-V2 / .agents/skills/deepworkplan/onboard
deepworkplan-onboard
VerifiedMake any repository AI-first — reason (never template) an adapted AGENTS.md, docs/, per-module docs and .agents/ kit from the real repo, discover and verify its full and scoped validation commands and source-to-test mapping, install the DeepWorkPlan skill, and, for a repository onboarded under an earlier version, perform a targeted, non-destructive, idempotent harness upgrade. Use when the developer wants to onboard or upgrade a repository for AI agents.
v6.md
View on GitHub# Bounded-autonomy onboarding guidance (v6 line)
> Load this file **only when** the repository will run v6 plans: the
> installed pack's line is 6+ or the developer explicitly requested the
> v6 (`../spec/V6_LIFECYCLE.md` §2). A v5-only repository never
> reads it, never gets its sections, and is never migrated implicitly.
> The normative text is [`../spec/V6_LIFECYCLE.md`](../spec/V6_LIFECYCLE.md)
> §2 and §10; the capability vocabulary is
> [`../spec/V6_RESOURCES.md`](../spec/V6_RESOURCES.md) §1.
Onboarding a v6 repository adds four records to the generated guidance.
Each is **reconciled non-destructively** (Phase 0 rules apply verbatim):
assess what exists, preserve equivalent meaning and every custom rule,
add only what is missing, and keep the `AGENTS.md` line budget. A second
pass with unchanged inputs must produce **no diff**.
## 1. Capability declaration — what the host can actually do
Record, in the repo's agent guidance, which of the eight closed-set
abilities the runtime host really provides: `stop_agent`, `meter_spend`,
`meter_tokens`, `meter_wall_clock`, `cancel_children`, `model_routing`,
`subagents`, `telemetry`.
- An unstated ability is `false` — never infer one from optimism.
- A resource limit the host cannot meter is **advisory, not enforced**;
the guidance names the missing ability rather than implying a bound.
- `telemetry` is **opt-in only**; a repository that has not consented
records `telemetry: false` and the flows never emit usage events.
- The minimal host (all abilities false) is a supported posture, never
an error and never a reason to refuse onboarding.
The declaration is a short prose block in the target repo (`AGENTS.md`
or an owner doc) **and** its machine-readable twin: record each verified
`true` with `python3 <pack>/shared/config.py host <capability> true
--repo .` (the `host` key of `.dwp/config.json`, `../spec/CONFIG.md`
§3a), which `shared/resources.py` reads at runtime. Prose alone is
invisible to the helpers.
## 2. Authority boundaries — who may authorize what
Record how authority works **in this repository**, from its real
approval rules found in Phase 1 reconnaissance — never boilerplate:
- who authors plans and who approves them (the v6 contract records the
mechanism: `plan_authorship` or `pre_authorization`, with the
authority named);
- the consent checkpoints an unattended run must stop at (pushes,
publication, external messages, secret access, anything the repo's own
rules reserve for a human);
- what existing authorization is reusable and must not be re-requested.
## 3. Outcome/test mapping — every outcome has an observable check
Record the repository's real outcome checks: the commands a v6 plan's
acceptance criteria and gate intents are drawn from. Use **the repo's
own commands discovered in reconnaissance** (`pnpm run test`,
`bats tests/`, `make check`, the fixture family's real runner) — never
an invented or aspirational command, and never a preset's suggestion
unverified against the repo. Keep the mapping where plan authors will
find it (usually the testing guidance doc), concise, and honest about
what is *not* covered by an automated check.
## 4. Concise working context — entry stays small
Apply the existing budget discipline to the v6 records: the entry point
(`AGENTS.md`) carries the capability block, the authority summary and
pointers; the detail lives in owner docs. Nothing here expands routine
per-task reads — v6 plans select context per task
(`../spec/V6_CONTEXT.md`), and onboarding must not undo that by
front-loading the harness.
## Upgrade scenarios
- **Partial harness upgrade** (repo onboarded by an older pack):
reconcile — never re-onboard from scratch. Preserve custom content and
existing plans; add the missing v6 records above as new sections; a
conflicting generated rule is reconciled, not duplicated.
- **The pack moved to the current v6 line**: existing v1/v2/v5
plans keep their recorded lifecycle. New plans follow the activation
rule (line 6+ or explicit request). Moving a v5 plan forward is the
explicit one-directional surface `../shared/migrate_v6.py` (preview
first, then `migrate --authority`, rollback available) — never a
silent conversion by onboarding or any other flow.
- **Second pass (idempotence)**: re-running onboarding with unchanged
repo inputs produces no diff — including for the four v6 records. If
a diff appears, one of the two passes is a defect.
## Cost honesty
When reporting this onboarding, measure and disclose **onboarding plus
first-work cost separately from amortized later work**. Setup is never
free in an end-to-end claim, the v6 detail is trigger-gated (a v5-only
repo pays nothing for this file), and no blanket model-performance
claim may be drawn from these numbers.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| finishing-a-development-branchobra/superpowersUse when implementation is complete, all tests pass, and you need to decide how to integrate the work | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| systematic-debuggingobra/superpowersUse when encountering any bug, test failure, or unexpected behavior, before proposing fixes | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| test-driven-developmentobra/superpowersUse when implementing any feature or bugfix, before writing implementation code | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| skill-creatoranthropics/skillsCreate new skills, modify and improve existing skills, and measure skill performance. Use when users want to create a skill from scratch, edit, or optimize an existing skill, run evals to test a skill, benchmark skill performance with variance analysis, or optimize a skill's description for better triggering accuracy. | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |