IFT-TS Roadmaps and Reports

AI Tooling

3 min read

ift-ts:web:ift:2026q4-ai-tooling

Description

Three tools are in scope: ai-review, already consumed by status-im/status-web, logos-co/logos-web, and acid-info/admin-acid.logos.co; an ai-auto-fix companion that turns accepted ai-review findings into a patch, not yet started; and ai-docs-sync, in development. All three ship as reusable GitHub workflows pinned to @v1, with zero npm dependencies, so a tool or model change is one edit in the shared repository rather than one per consumer. Human approval before merge stays a hard requirement for every tool here.

Task List

ai-review maintenance and tuning

  • fully qualified name: ift-ts:web:ift:2026q4-ai-tooling:ai-review
  • owner: JulesFiliot
  • status: not started
  • start-date: 2026/10/01
  • end-date: 2026/12/31

Description

Maintain acid-info/ai-review as the shared PR reviewer for logos-co, status-im, and acid-info repositories. Keep the dual-model review plus synthesis pipeline reliable as provider APIs and models change, tune per-repository context and guideline files to cut false positives, and refine diff packing, path filtering, and severity thresholds. Keep the degraded paths honest: a single-model review when one provider is down, a local merge when synthesis fails, and a posted body that says the review was partial. Onboard remaining Web repositories and keep the @v1 tag stable for consumers.

Deliverables

  • PRs in acid-info/ai-review for model updates, prompt and threshold tuning, and pipeline fixes.
  • Repository-specific guideline and config files in consumer repositories.
  • Additional consumer repositories onboarded via .github/workflows/ai-review.yml.
  • A short assessment of signal versus noise, drawn from reviewed PRs, with the tuning changes it drove.

ai-auto-fix companion

  • fully qualified name: ift-ts:web:ift:2026q4-ai-tooling:ai-auto-fix
  • owner: JulesFiliot
  • status: not started
  • start-date: 2026/10/01
  • end-date: 2026/12/31

Description

Build an ai-auto-fix companion to ai-review that takes review findings and produces a patch the author can accept, rather than leaving every finding as manual follow-up work. Nothing exists yet, so the task starts from specification: which finding severities and categories are eligible, how a fix is requested (a comment command on the PR, or opt-in per finding), how the patch is delivered (a commit pushed to the PR branch, or a suggested-change comment), and what the tool must refuse to touch. Reuse the ai-review pipeline’s shape — reusable workflow, zero npm dependencies, Node 22 fetch, pinned @v1 — and keep every produced change inside normal PR review with no branch-protection bypass.

Deliverables

  • Written specification covering eligible findings, trigger mechanism, patch delivery, and refusal rules.
  • Prototype workflow producing patches against real ai-review findings on Web repositories.
  • Documented guardrails: disallowed change types, secret handling, and required human approval before merge.

ai-docs-sync

  • fully qualified name: ift-ts:web:ift:2026q4-ai-tooling:ai-docs-sync
  • owner: JulesFiliot
  • status: not started
  • start-date: 2026/10/01
  • end-date: 2026/12/31

Description

Finish acid-info/ai-docs-sync, which keeps a repository’s documentation in step with its code by triaging each push, rewriting affected docs with one model, checking the rewrite with a second, and running mechanical gates.

Deliverables

  • Merged publishing stage: rolling branch and PR, cursor move, generated PR body, with unit tests.
  • End-to-end runs on a real repository, including unmerged-edit carry-forward and collapsed-push handling.
  • v1 tag cut and the README status banner removed.
  • First consumer repository onboarded via .github/workflows/docs-sync.yml, with the resulting docs PRs linked.

Graph View