fix: list_resources carries the TUI's status columns — no more N+1 get_resource

[hellices]

Closes #158

Problem

_list_resources rendered one shape for every kind — namespace/name - age=… — dropping every status fact the summaries carry. The observed cost: an MCP host asked to "check pod status" got status-less rows, then issued N+1 get_resource calls (full masked manifests, token-expensive) to reconstruct what the TUI table already showed; worse, on the pods path the data was dropped twice (summary_for("Pod") fell through to GenericSummary, capturing nothing), so the host could confidently report pods "fine" while the screen showed CrashLoopBackOff.

Fix

1. Pod-aware LIST summary — summary_for dispatches Pod to a new PodListSummary(GenericSummary): kubectl-style display STATUS (waiting reasons like CrashLoopBackOff included, via the existing _display_phase), READY n/m, RESTARTS, NODE. Tool-path only: the watch path's richer PodSummary and the resource store are untouched.

2. Kind-aware facts, type-keyed — list_resources lines now mirror the TUI columns per kind:

Kind	Line facts
pods	phase=… ready=n/m restarts=k node=…
replicasets	revision=… desired=… current=… ready=…
subscriptions / CSVs / packagemanifests	the same facts list_operators prints
generic kinds	desired=n when spec.replicas exists
helm release/revision	renderers registered ahead of the listing tool (#161)

3. Custom columns (issue #45) — user-configured columns reach the model as NAME=value (clamped to 80 chars/value so a hostile value cannot dominate the result budget). ToolExecutor takes the configured column names per plural — same source as KubeClient's value wiring — injected at all three construction sites in __main__.py.

4. Drift contract — the renderer table is the contract: test_every_typed_summary_has_a_facts_renderer walks GenericSummary.__subclasses__() and fails when any typed summary lacks a facts renderer, so a future kind cannot silently degrade back to name+age. (It already caught HelmReleaseSummary/HelmRevisionSummary during development.)

Split out: #161 (helm_list_releases read tool — needs a new MCP-surface tool + follow-mode pairing, its own review cycle).

Testing

tests/tools/test_list_resources.py (new, 10 tests): Pod summary capture incl. CrashLoopBackOff-as-status, per-kind facts lines, generic desired/empty cases, the subclass contract, end-to-end tool output for pods and custom columns (incl. clamping). Wiring pinned in tests/test_main_wiring.py::test_mcp_executor_receives_custom_column_names.

Full gate green: ruff, mypy --strict, tach, 2966 passed / 21 skipped, coverage ≥ 80%.

[Copilot]

Pull request overview

Adds status-rich, kind-aware list_resources output to avoid N+1 manifest fetches.

Changes:

• Adds pod and typed-summary fact rendering.
• Exposes configured custom columns with value clamping.
• Wires column names into agent and MCP executors with tests.

Reviewed changes

Copilot reviewed 5 out of 5 changed files in this pull request and generated 1 comment.

Show a summary per file

File	Description
src/korvid/tools/executor.py	Renders kind-specific and custom facts.
src/korvid/k8s/models.py	Adds pod-aware list summaries.
src/korvid/__main__.py	Wires custom-column names.
tests/tools/test_list_resources.py	Tests summary and output behavior.
tests/test_main_wiring.py	Tests MCP column wiring.

Suppressed comments (2)

src/korvid/tools/executor.py:104

• PackageManifestSummary.description is rendered in the TUI (resource_table.py:750-756) but is omitted here, so list_resources(kind="packagemanifests") still violates this PR's column-parity goal. Include a clamped description when present and assert it in the package facts test.

def _package_facts(s: PackageManifestSummary) -> str:
    return (
        f"catalog={s.catalog or '?'} default_channel={s.default_channel or '?'}"
        f" channels={_clamp(','.join(s.channels)) or '?'}"
    )

src/korvid/tools/executor.py:123

• The Helm revision TUI includes APP VERSION (resource_table.py:709-716), but this registered renderer drops s.app_version. When the follow-up exposes this renderer, revision output will already drift from the screen; add the clamped field and cover the Helm renderers in the contract tests.

        f"chart={_clamp(s.chart)}",

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[my-reviewer-agent]

이슈 #158 해결 확인 — pods LIST 경로가 GenericSummary로 떨어져 상태가 두 번 사라지던 문제를 PodListSummary(기존 _display_phase 재사용, TUI PodSummary.from_manifest와 phase/ready/restarts/node 계산 로직 동일함을 head에서 대조 확인)로 닫았고, kind별 facts 렌더러 + __subclasses__() 드리프트 계약 테스트, 커스텀 컬럼 name=value(80자 클램프) 배선(main.py 3개 구성 지점 모두 확인)까지 설계가 깔끔합니다. CrashLoopBackOff-as-status 테스트가 이슈의 헤드라인 실패를 정확히 고정합니다. 인라인 2건은 비차단(개행 주입 Warning 1건, 계약 테스트 사각 Suggestion 1건)입니다.

APPROVE

[my-reviewer-agent]

재리뷰 (신규 커밋 1건: 47e96bb)

라운드 1의 Warning(커스텀 컬럼 값에 포함된 개행/제어문자로 model-facing 결과에 가짜 행 주입 가능)을 정확히 해결했습니다.

• _clamp가 길이 제한 전에 isprintable() 기반으로 비출력 문자를 공백으로 평탄화 — #156의 toast sanitizer와 동일한 접근이고, 모든 fact 경로(_pod_facts/generic/custom)가 _clamp를 경유하므로 커버리지 완전.
• 회귀 테스트 test_custom_column_values_cannot_forge_extra_rows가 실제 공격 형태(개행으로 위조한 pod 행 + ANSI \x1b[2J)를 사용하고 splitlines()==1 + \x1b 부재를 단언 — 형식적이지 않은 진짜 테스트입니다.

지적사항 없음. 라운드 1의 __subclasses__() 직계-한정 Suggestion은 여전히 열려 있으나 advisory 수준이므로 승인에 영향 없습니다.

APPROVE

[Copilot]

Pull request overview

Copilot reviewed 5 out of 5 changed files in this pull request and generated no new comments.

Suppressed comments (5)

src/korvid/tools/executor.py:101

• version and phase are raw CSV CRD strings and remain unsanitized, unlike display_name. A crafted value can therefore add fake result rows or dominate the capped output; clamp every cluster-derived string in this renderer.

    parts = [f"version={s.version or '?'}", f"phase={s.phase or '?'}"]
    if s.display_name:
        parts.append(f"display={_clamp(s.display_name)}")

src/korvid/tools/executor.py:95

• These Subscription fields are cluster-controlled CRD strings, but none passes through _clamp. In particular, a user-supplied channel can contain an embedded newline or be arbitrarily long, reintroducing forged rows/result-budget exhaustion on this typed path.

    return (
        f"channel={s.channel or '?'} source={s.source or '?'}"
        f" csv={s.installed_csv or '?'} state={s.state or '?'}"
    )

src/korvid/tools/executor.py:109

• catalog and default_channel are also cluster-derived strings, but only channels is clamped. Catalog data containing controls/newlines or a very long value can still forge rows or consume the result budget.

    return (
        f"catalog={s.catalog or '?'} default_channel={s.default_channel or '?'}"
        f" channels={_clamp(','.join(s.channels)) or '?'}"
    )

src/korvid/tools/executor.py:88

• revision is read from the freely writable deployment.kubernetes.io/revision annotation, so it can contain newlines/control characters or an oversized value. Because it is emitted without _clamp, one ReplicaSet can forge extra rows or crowd out the real listing despite the new sanitization contract.

    return f"revision={s.revision} desired={s.desired} current={s.current} ready={s.ready}"

src/korvid/tools/executor.py:81

• phase comes from pod status/waiting-reason strings and bypasses _clamp, so a malformed or hostile status can still inject a newline/control sequence into this model-facing list or consume the result budget. Apply the same single-line bounded rendering used for node and custom columns.

This issue also appears in the following locations of the same file:

• line 88
• line 92
• line 99
• line 106

    parts = [f"phase={s.phase or '?'}", f"ready={s.ready or '?'}", f"restarts={s.restarts}"]

[Copilot]

Pull request overview

Copilot reviewed 5 out of 5 changed files in this pull request and generated no new comments.

Suppressed comments (1)

src/korvid/tools/executor.py:137

• _helm_revision_facts drops app_version, even though the Helm revision table includes APP VERSION (resource_table.py:57,714) and the #161 contract names it as part of this renderer. Once the follow-up exposes this renderer, revision results will still disagree with the TUI. Add the clamped app version alongside the chart.

        f"chart={_clamp(s.chart)}",

[hellices]

Round 3's suppressed finding is fixed: _helm_revision_facts now includes the clamped app_version alongside chart, matching the TUI's revision table (test_helm_revision_facts_include_app_version). Full gate green (2970 passed).

[my-reviewer-agent]

재리뷰 (신규 커밋 2건) — 이전 라운드의 두 가지 잔여 사항이 모두 해결되었습니다.

1. 클러스터 유래 문자열 전면 클램프: phase/revision(자유 기록 가능한 annotation)/channel/source/csv/state/catalog/default_channel/version/helm status 등 모든 렌더러가 _clamp(비인쇄문자 평탄화 + 길이 제한)를 경유하도록 확장. test_every_cluster_derived_string_is_flattened가 개행 주입 페이로드로 5개 summary 타입을 실검증 — 이전 라운드의 부분 적용 문제를 완전히 닫음.
2. MRO 기반 renderer dispatch: 이전에 남긴 __subclasses__() direct-only Suggestion 해결. summary_facts가 type(s).__mro__를 순회해 손자 서브클래스도 부모 렌더러를 상속하고, contract 테스트도 서브클래스 트리를 재귀 순회(test_grandchild_summary_uses_its_parents_renderer)로 검증.
3. helm revision facts에 app_version 추가: TUI의 APP VERSION 컬럼과 1:1 일치, 테스트 포함.

지적사항 없음. APPROVE

[Copilot]

Pull request overview

Copilot reviewed 5 out of 5 changed files in this pull request and generated no new comments.