fabric drift validates only declared edges — card-vs-source divergence is undetectable (37% of cards affected here)

Đang mở
  • Có gì không ổn?
#70 0 bình luận 0 reaction 0 người được giao Xem trên GitHub

Chưa có ai nhận issue này.

Đánh giá

Độ khó
5/5
Thời gian dự kiến
Hơn một tuần
Mức phù hợp với người mới
35/100
Loại issue
Tính năng
Độ rõ ràng
Khá rõ ràng
Mức độ hoạt động
Sôi nổi
Công nghệ
shell
Lĩnh vực
cli, devtools

Hướng nghiên cứu

Start with agents/fabric/lib/drift.sh and trace the stale-edge check, then inspect enrich.py and its detect_ts_js_imports detector to understand how declared and source-derived edges are represented. Run the detector and fw fabric drift on a representative card/source pair, including missing and extra edges. Done means drift reports card/source divergence and tests cover both directions; severity and coverage-ratchet behavior need an upstream design decision.

Do mô hình lập chỉ mục viết ra từ nội dung của issue.

Mô tả

Context

  • Framework version: 1.6.258
  • Project: 050-email-archive (shared-tooling mode)
  • Reported: 2026-09-25T06:34:48Z

Description

Field report from 050-email-archive (Pen). Local task T-2245. Surfaced by a real 185-day outage (T-2243).

THE INVARIANT THAT IS NOT HELD

agents/fabric/lib/drift.sh's stale-edge check asks, for each edge a card DECLARES, whether the target resolves on disk (_resolves_on_disk). Nothing asks whether the card's declared edges still match what the source actually imports. An import present in the code and absent from the card is invisible to drift. So a card can be arbitrarily incomplete — or frozen years ago — and 'fw fabric drift' reports clean indefinitely.

CLAUDE.md describes the Component Fabric as 'the source of truth for structural relationships'. It cannot currently detect that it has drifted from the code it maps.

MEASURED HERE (2026-09-25)

Using the framework's OWN detector — enrich.py detect_ts_js_imports — run against the same source file each card describes:

240 of 643 JS cards declare fewer edges than their source actually imports.
fw fabric drift reports 'unregistered: 0, orphaned: 0, stale: 0' across all of them.

Spot-checked to rule out a path-form artifact: scripts/_add-jp-to-cohort-events.js declares exactly one edge (microsoftTokenService.js) and contains zero mention of src/database/models in any form, while the source genuinely requires it (resolves via index.js). The divergence is real, not a spelling mismatch.

WHAT IT COST

Commit da56fd12 moved four service files into services/legacy/ as pure renames with no content change and updated no require path. Three whole-archive import entry points threw MODULE_NOT_FOUND on their first require for 185 days. All three HAD fabric cards WITH depends_on edges — but none of those edges named the importer that broke, and the declared targets (config files) still existed, so drift stayed green the entire time. Card last_verified: 2026-03-18, six days before the breaking commit.

THIS IS NOT A DETECTOR BUG

detect_ts_js_imports is correct and already shipped (T-552, added precisely because a TS project showed 0 edges). Run directly on the broken file it returns the right answer. The gap is that 'enrich' is a manual step with no staleness signal, and 'drift' — the verb whose job is detecting drift — does not include card-vs-source divergence in what it checks.

Strong suspicion this is also the root of the recurring 'N of M fabric cards have no edges' audit warning: same cause, extreme case.

SUGGESTED FIX

Add a fourth section to 'fw fabric drift', alongside unregistered / orphaned / stale: divergent — cards whose declared edges differ from what the enrich detectors derive from the same source. The detectors already exist and already work; this is wiring them into the read path rather than only the write path. Effectively 'enrich --dry-run' surfaced as drift.

Two design points worth deciding upstream rather than by a consumer:

  1. Direction. Missing edges (code imports X, card omits it) are the dangerous case. Extra edges (card declares X, code no longer imports it) are also drift but lower severity — the existing stale check catches only the subset where X was also deleted from disk.
  2. Severity. If divergence were WARN rather than FAIL, a project like this one would start at 240 findings. A coverage-percentage metric with a ratchet may be more usable than a per-card finding.

NOT PATCHED LOCALLY. A worker was running concurrently in this project and the governing mandate forbids opening a structural change alongside one. Reporting first, patching after.

Reported via fw upstream report from field installation

Ngôn ngữ chính
Shell
Star
16
Fork
6
Không có pull request nào được merge trong 30 ngày

Chuẩn bị môi trường

Bắt đầu từ đâu

  1. Đọc hết issue, rồi đọc hướng dẫn đóng góp của dự án.
  2. Bình luận trên issue rằng bạn sẽ nhận — tránh hai người làm cùng một việc.
  3. Fork repository và làm thay đổi trên một nhánh.
  4. Mở pull request có tham chiếu số hiệu của issue.

Issue khác của DimitriGeelen/agentic-engineering-framework

Tất cả issue của DimitriGeelen/agentic-engineering-framework

Issue tương tự

Thêm issue về Shell/Bash

Nhận issue mới trong hộp thư của bạn

Bản tóm tắt ngắn những issue GitHub phù hợp với người mới.