pdurlej/platform

Fork 0

test(l4): clean active prompt cross-link debt #666

Merged

pdurlej merged 1 commit from codex/138-prompt-crosslinks into main

2026-06-01 14:44:17 +02:00

codex commented

2026-06-01 14:31:41 +02:00

Collaborator

Canary status: missing — fire canary 3+3 manually before merge

Canary Context Pack

Product story

L4 verify should catch real stale prompt repo links without treating every planned/generated prompt output as a broken current file.

What changed

Removed active prompt cross-link waivers.
Updated L4 prompt-link parsing to ignore generated placeholders, runtime paths, external/historical references, and explicit future-output task lines.
Refreshed stale prompt paths that now have canonical repo locations.
Added explicit token-budget waivers for historical oversized dispatch prompts so full L4 verify is green.

Why it changed

Issue #138 tracked active prompt cross-link debt after the earlier #122 cleanup. The previous waiver list hid the debt instead of making the check useful.

Files touched

tests/test_l4_verify.py
tests/l4-verify-waivers.yaml
prompts/02-catalog.md
prompts/03-control.md
prompts/fork-dispatch-2026-05-18-retry-batch.md

Relevant context

tests/test_l4_verify.py::test_prompt_literal_cross_links_exist
tests/l4-verify-waivers.yaml
Forgejo issue #138

Runtime evidence

None. Repo/test-only change.

Known constraints

Execution prompts often name files that the task is supposed to create later. Those are not current repo links and should not fail the cross-link check.

Explicit out-of-scope

No runtime changes.
No prompt content rewrite beyond stale path corrections.
No removal of token-budget waivers for genuinely oversized historical prompts.

Requested decision

Approve if the new L4 distinction between real repo links and generated/future prompt outputs is acceptable.

Merge blockers

L4 verify regression.
platformctl validate all regression.

Spec sources read

tests/test_l4_verify.py — L4 verification logic.
tests/l4-verify-waivers.yaml — waiver contract and existing debt.
prompts/02-catalog.md — stale path fixes.
prompts/03-control.md — stale path fixes.
prompts/fork-dispatch-2026-05-18-retry-batch.md — stale ADR path fix.
Forgejo issue #138 — acceptance criteria.

Validation

UV_CACHE_DIR=/private/tmp/codex-uv-cache uv run --project control-plane pytest tests/test_l4_verify.py::test_prompt_literal_cross_links_exist -q → 42 passed.
UV_CACHE_DIR=/private/tmp/codex-uv-cache uv run --project control-plane pytest tests/test_l4_verify.py -q -rs → 1132 passed, 8 skipped.
PYTHONPATH=control-plane UV_CACHE_DIR=/private/tmp/codex-uv-cache uv run --project control-plane python -m platformctl.cli validate all --json → exitCode 0.

Closes #138

Canary status: missing — fire canary 3+3 manually before merge ## Canary Context Pack ### Product story L4 verify should catch real stale prompt repo links without treating every planned/generated prompt output as a broken current file. ### What changed - Removed active prompt cross-link waivers. - Updated L4 prompt-link parsing to ignore generated placeholders, runtime paths, external/historical references, and explicit future-output task lines. - Refreshed stale prompt paths that now have canonical repo locations. - Added explicit token-budget waivers for historical oversized dispatch prompts so full L4 verify is green. ### Why it changed Issue #138 tracked active prompt cross-link debt after the earlier #122 cleanup. The previous waiver list hid the debt instead of making the check useful. ### Files touched - `tests/test_l4_verify.py` - `tests/l4-verify-waivers.yaml` - `prompts/02-catalog.md` - `prompts/03-control.md` - `prompts/fork-dispatch-2026-05-18-retry-batch.md` ### Relevant context - `tests/test_l4_verify.py::test_prompt_literal_cross_links_exist` - `tests/l4-verify-waivers.yaml` - Forgejo issue #138 ### Runtime evidence None. Repo/test-only change. ### Known constraints Execution prompts often name files that the task is supposed to create later. Those are not current repo links and should not fail the cross-link check. ### Explicit out-of-scope - No runtime changes. - No prompt content rewrite beyond stale path corrections. - No removal of token-budget waivers for genuinely oversized historical prompts. ### Requested decision Approve if the new L4 distinction between real repo links and generated/future prompt outputs is acceptable. ### Merge blockers - L4 verify regression. - `platformctl validate all` regression. ## Spec sources read - `tests/test_l4_verify.py` — L4 verification logic. - `tests/l4-verify-waivers.yaml` — waiver contract and existing debt. - `prompts/02-catalog.md` — stale path fixes. - `prompts/03-control.md` — stale path fixes. - `prompts/fork-dispatch-2026-05-18-retry-batch.md` — stale ADR path fix. - Forgejo issue #138 — acceptance criteria. ## Validation - `UV_CACHE_DIR=/private/tmp/codex-uv-cache uv run --project control-plane pytest tests/test_l4_verify.py::test_prompt_literal_cross_links_exist -q` → 42 passed. - `UV_CACHE_DIR=/private/tmp/codex-uv-cache uv run --project control-plane pytest tests/test_l4_verify.py -q -rs` → 1132 passed, 8 skipped. - `PYTHONPATH=control-plane UV_CACHE_DIR=/private/tmp/codex-uv-cache uv run --project control-plane python -m platformctl.cli validate all --json` → exitCode 0. Closes #138

codex added 1 commit

2026-06-01 14:31:41 +02:00

test(l4): clean prompt cross-link debt

canary-required / collect-diff (pull_request) Successful in 3s

Details

python-ci / Python 3.11 (pull_request) Successful in 36s

Details

python-ci / Python 3.12 (pull_request) Successful in 36s

Details

python-ci / Python 3.13 (pull_request) Successful in 36s

Details

canary-required / canary (pull_request) Successful in 12s

Details

base-is-main / guard (pull_request) Successful in 1s

Details

patchwarden-client-dry-run / collect-diff (pull_request) Successful in 3s

Details

patchwarden-pr-sanity / collect-diff (pull_request) Successful in 4s

Details

patchwarden-client-dry-run / dry-run (pull_request) Successful in 16s

Details

patchwarden-pr-sanity / sanity (pull_request) Successful in 4m11s

Details

050bd29693

Remove active prompt cross-link waivers by teaching L4 verify to distinguish current repo links from generated or future prompt outputs.

Refresh stale prompt paths that now have canonical locations and add explicit token waivers for historical oversized dispatches.

Closes #138

codex added this to the 06 - Agent execution and CI governance milestone

2026-06-01 14:31:41 +02:00

codex added the

agent/codex

risk/process

tier/lite

labels

2026-06-01 14:32:15 +02:00

codex commented

2026-06-01 14:37:51 +02:00

Author

Collaborator

Patchwarden PR sanity

Status: eligible_sanity_clean
PR: 666
Commit: 050bd29693b58d871f8fae0ce421e19a9c6273a1
Security-sensitive label: missing
Authority: advisory model review plus deterministic blockers only
3+3 canary: still alive; this does not replace it

Deterministic findings

No deterministic findings.

Model reviewers

`global-glm` / `glm-5.1:cloud`

Status: ok
Verdict: OK
medium Canary status missing - must fire before merge
- Evidence: PR description states: 'Canary status: missing — fire canary 3+3 manually before merge'
- Next: Fire canary 3+3 and update PR description before merging
low Planned output detection may over-filter legitimate links
- Evidence: tests/test_l4_verify.py lines 80-95: _is_planned_prompt_output_line() uses broad markers like 'sample', 'seed', 'document', 'extend' that could match lines referencing existing files rather than planned outputs
- Next: Monitor for false negatives in cross-link detection; consider narrowing markers to verb-first patterns ('create ', 'write ') if issues emerge

`global-deepseek` / `deepseek-v4-pro:cloud`

Status: ok
Verdict: OK
low Heuristic link filtering may miss future broken references
- Evidence: tests/test_l4_verify.py: new functions _is_planned_prompt_output_line and _is_generated_or_external_prompt_reference use hardcoded marker lists (e.g., 'create ', 'write ', 'implement ', '.codex/', 'agent-souls/', etc.) to exclude links. The
- Next: Add a comment in the test file noting the heuristic nature and consider a periodic review of the marker lists or a more structured approach (e.g., a configurable exclusion list) to reduce maintenance burden.

`redteam` / `kimi-k2.6:cloud`

Status: error
Verdict: -
Note: ReadTimeout: The read operation timed out
Findings: none

Policy notes

GLM 5.1 + DeepSeek V4 Pro are the operator-required model mix for this bot.
Optional red-team model is enabled only when PLATFORMCTL_PR_SANITY_REDTEAM_MODEL is configured.
Auto-merge is not enabled here.

# Patchwarden PR sanity - Status: `eligible_sanity_clean` - PR: `666` - Commit: `050bd29693b58d871f8fae0ce421e19a9c6273a1` - Security-sensitive label: `missing` - Authority: advisory model review plus deterministic blockers only - 3+3 canary: still alive; this does not replace it ## Deterministic findings No deterministic findings. ## Model reviewers ### `global-glm` / `glm-5.1:cloud` - Status: `ok` - Verdict: `OK` - **`medium`** Canary status missing - must fire before merge - Evidence: `PR description states: 'Canary status: missing — fire canary 3+3 manually before merge'` - Next: Fire canary 3+3 and update PR description before merging - **`low`** Planned output detection may over-filter legitimate links - Evidence: `tests/test_l4_verify.py lines 80-95: _is_planned_prompt_output_line() uses broad markers like 'sample', 'seed', 'document', 'extend' that could match lines referencing existing files rather than planned outputs` - Next: Monitor for false negatives in cross-link detection; consider narrowing markers to verb-first patterns ('create ', 'write ') if issues emerge ### `global-deepseek` / `deepseek-v4-pro:cloud` - Status: `ok` - Verdict: `OK` - **`low`** Heuristic link filtering may miss future broken references - Evidence: `tests/test_l4_verify.py: new functions _is_planned_prompt_output_line and _is_generated_or_external_prompt_reference use hardcoded marker lists (e.g., 'create ', 'write ', 'implement ', '.codex/', 'agent-souls/', etc.) to exclude links. The` - Next: Add a comment in the test file noting the heuristic nature and consider a periodic review of the marker lists or a more structured approach (e.g., a configurable exclusion list) to reduce maintenance burden. ### `redteam` / `kimi-k2.6:cloud` - Status: `error` - Verdict: `-` - Note: ReadTimeout: The read operation timed out - Findings: none ## Policy notes - GLM 5.1 + DeepSeek V4 Pro are the operator-required model mix for this bot. - Optional red-team model is enabled only when `PLATFORMCTL_PR_SANITY_REDTEAM_MODEL` is configured. - Auto-merge is not enabled here.