Files
LumaWeb/state/evidence/S00-012-manual-ci-cd-stdin-20261015b.md
T
wt c28412ae8f S00-012: Record blocked task and remediation handoff
Changes: Preserve incomplete work and blocker metadata for S00-012. This checkpoint does not mark implementation accepted.
Validation: Session ended with status blocked.
Blocker: Required automation suite fails in existing all-prompts budget test: S21-006 exceeds even a diagnostic 90000-character prompt limit. This blocks verification of the stdin remediation; production limit remains 80000.

Task: S00-012
Agent: manual-ci-cd-stdin-20261015b
Role: ci-cd
Action: escalate
Status: blocked

Task title: Fix runner prompt delivery after partial stdin timeout
2026-10-15 01:37:36 +07:00

2.8 KiB

S00-012 partial implementation and blocker evidence

Executor: manual-ci-cd-stdin-20261015b (ci-cd). No submit, review acceptance or done is claimed. Input: S00-012 card/context packet, diagnostic evidence, accepted ADR 0000 and reviewed S00-001/S00-011 dependency results. Output attempted: complete stdin prompt delivery under startup delay and pipe backpressure. No product/shared contract changes required.

Actual changes in scope

scripts/runner.py now opens the already-persisted UTF-8 prompt.md as binary stdin for the child and polls process.wait rather than attempting timed partial pipe writes through communicate. EOF and complete prompt bytes are supplied by the file; .2s state monitoring, deadlines and stop_process_tree cleanup remain. No prompt truncation or limit changes.

tests/tooling/test_automation.py adds a delayed-reader regression: synthetic tool harness waits .5s before reading 140024 UTF-8 bytes (emoji and Cyrillic), compares all bytes with LUMA_PROMPT_FILE, observes EOF, then submits one fixture task. No real model/provider task is run.

Actual commands/results

  • tasks.py resume: exit 0, empty owned tasks; claim S00-012: exit 0, revision 13; show/packet read.
  • Before runner fix: python3 -m unittest discover -s tests/tooling -p test_automation.py -k delayed_reader -v: exit 1. Custom load_tests runs all 23 automation tests despite -k. New delayed-reader regression failed (blocked instead of review) and existing all-prompts budget test errored.
  • After runner fix: python3 -m unittest discover -s tests/tooling -p test_automation.py -v: exit 1, 23 tests in 4.465s. New delayed-reader regression passed; 22 tests passed, one existing FoundationRunnerTests.test_all_current_fresh_agent_prompts_fit_configured_budget errored. Timeout, scope and terminal-transition cleanup tests passed. Full suite is not passed.
  • Read-only diagnostic iteration with max_prompt_chars=90000: exit 0; only S21-006 still raises Prompt exceeds configured character budget. Production configuration remains 80000 characters; raising it to the allowed maximum still does not fit S21-006.
  • git diff --stat scoped inspection: exit 0; only scripts/runner.py and tests/tooling/test_automation.py source files changed by this executor. Pre-existing uncommitted S00-012 blocker/workflow artifacts were present at entry and preserved.

Blocker and limits

Existing prompt-budget validation fails on S21-006 independently of stdin transport. Required tooling suite cannot pass. Stop source edits and create one separate remediation for bounded context preparation; do not truncate requirements, weaken validation or change immutable specs. make check was not run/passed after this blocker was established. No real Codex launch was attempted. Python runtime was 3.9.6. No Git mutations by executor; partial source/workflow changes remain uncommitted for coordinator checkpoint.