Plan minimal d’implémentation après spike Paru #21

New issue

Closed

opened 2026-08-31 03:00:46 +00:00 by hephaistos · 0 comments

hephaistos commented

2026-08-31 03:00:46 +00:00

Owner

MODE: EXECUTION

Objective

After PR #20 is merged, prove the one unresolved Paru integration boundary in disposable Arch and write the short implementation plan for the accepted minimal LLM-first AURoscope architecture.

Definition of Ready

PR #20 is merged into current origin/main;
AGENTS.md, README.md, docs/specification.md, docs/architecture/minimal-v1.md, docs/design-phase.md, and ADR-0015 are present on main;
the work starts from a clean branch based on that origin/main;
no production implementation from the superseded architecture is reused.

Authorized scope

Run a narrow disposable-Arch spike that proves, for one explicitly supported Paru version:
- native search/selection and AUR origin identification;
- exact AUR worktree availability before package-supplied code;
- whether edit → re-audit → final build can use the same worktree;
- minimal PreBuildCommand timing and identity check;
- skip/exclusion followed by a fresh native Paru resolution;
- official repository update/install paths remain audit-free and transparent.
Record durable non-obvious evidence under docs/dependency-notes/.
Write a new concise docs/implementation/initial-plan.md with vertical implementation steps centered on:
- transparent Paru/Pacman wrapper behavior;
- exact AUR recipe bundle and differential baseline;
- mandatory Codex CLI audit;
- per-pkgbase decisions;
- final native Paru resolution/build/install;
- minimal SQLite state and identity guard;
- proportionate tests and one disposable-Arch E2E.
If the spike cannot prove edit reuse cleanly, defer edit in the plan rather than taking over Paru's build role.
Open a Forgejo PR containing the evidence and plan.

Excluded scope

no production Go implementation;
no implementation of the wrapper, Codex adapter, UI, database, or guard;
no deterministic scanner, HTTP backend, resolver replacement, broad state machine, or autonomous implementation loop;
no merge or release.

Completion criteria

the spike answers the exact worktree/edit/guard/final-Paru question with reproducible commands and observed output;
the new plan is short, executable by a fresh agent, and contains no architecture beyond ADR-0015;
every plan step has a concrete outcome and verification gate;
active docs contain no reference to the superseded PR #18 plan as implementation guidance;
branch is pushed and the planning PR is ready for Mathieu's review.

MODE: EXECUTION ## Objective After PR #20 is merged, prove the one unresolved Paru integration boundary in disposable Arch and write the short implementation plan for the accepted minimal LLM-first AURoscope architecture. ## Definition of Ready - PR #20 is merged into current `origin/main`; - `AGENTS.md`, `README.md`, `docs/specification.md`, `docs/architecture/minimal-v1.md`, `docs/design-phase.md`, and ADR-0015 are present on `main`; - the work starts from a clean branch based on that `origin/main`; - no production implementation from the superseded architecture is reused. ## Authorized scope 1. Run a narrow disposable-Arch spike that proves, for one explicitly supported Paru version: - native search/selection and AUR origin identification; - exact AUR worktree availability before package-supplied code; - whether edit → re-audit → final build can use the same worktree; - minimal `PreBuildCommand` timing and identity check; - skip/exclusion followed by a fresh native Paru resolution; - official repository update/install paths remain audit-free and transparent. 2. Record durable non-obvious evidence under `docs/dependency-notes/`. 3. Write a new concise `docs/implementation/initial-plan.md` with vertical implementation steps centered on: - transparent Paru/Pacman wrapper behavior; - exact AUR recipe bundle and differential baseline; - mandatory Codex CLI audit; - per-`pkgbase` decisions; - final native Paru resolution/build/install; - minimal SQLite state and identity guard; - proportionate tests and one disposable-Arch E2E. 4. If the spike cannot prove edit reuse cleanly, defer `edit` in the plan rather than taking over Paru's build role. 5. Open a Forgejo PR containing the evidence and plan. ## Excluded scope - no production Go implementation; - no implementation of the wrapper, Codex adapter, UI, database, or guard; - no deterministic scanner, HTTP backend, resolver replacement, broad state machine, or autonomous implementation loop; - no merge or release. ## Completion criteria - the spike answers the exact worktree/edit/guard/final-Paru question with reproducible commands and observed output; - the new plan is short, executable by a fresh agent, and contains no architecture beyond ADR-0015; - every plan step has a concrete outcome and verification gate; - active docs contain no reference to the superseded PR #18 plan as implementation guidance; - branch is pushed and the planning PR is ready for Mathieu's review.

Rows
Columns