Reasoning models at high effort could spend the entire max_tokens budget
on thinking and return zero visible content with finish_reason=length;
the empty article flowed writer -> editor (reject) -> empty draft post,
with the cost booked as success (2026-06-11 incident: prod run acf24029,
draft post 921; diagnosis: convo pipeline-v2-phase1 seq 12).
- Empty-completion guard at the provider seam (new
OpenRouter_Completion_Guard): empty/whitespace content is a failure;
one retry with reasoning disabled (per-call override) when
finish_reason=length or reasoning was active, then the stage fails.
- finish_reason=length always warning-logged (model, stage, tokens).
- Failed/retried attempts booked to generation_log with
response_status=error + real token burn; run_decisions verdict rows
(retry_reasoning_disabled / failed_empty_completion) for v0.18.0 runs.
- Writer-path guard in Content_Generator::execute_stage and
belt-and-braces empty check in Publisher::create_post.
- Reasoning budget sanity: reasoning.max_tokens cap + completion
headroom via new prautoblogger_reasoning_max_tokens setting (default
constant 2048, 0 = legacy uncapped effort mode); cap replaces effort
when active; governor reservations use the effective ceiling.
- Body assembly moved to Request_Builder::build_body (300-line cap).
- Tests: guard verdict matrix, builder budget shaping, writer/publisher
guards, full provider retry path.
Agent-Session: prab-empty-draft-guard-20260611
Pipeline v2 Phase 1, commit 2/4. Prompt_Registry (read) +
Prompt_Registry_Writer (immutable create/activate/seed) over the new
prompts table; canonical v1 bodies extracted verbatim into
Prompt_Defaults(+Editorial) — single source for BOTH the seed migration
(Migrate_Prompt_Seed_V0180, flag-gated, self-healing) and the in-code
fallback, so a missing table renders byte-identical prompts and never
fatals. Call sites swapped: Content_Prompts, Analysis_Prompts,
Chief_Editor, LLM_Research_Provider. Includes the image-composer seam
keys image.rewriter_system + image.style_template (composer PR consumes,
must not add its own prompt storage).
Pinning: Cost_Tracker::set_run_id() registers the process run context,
ensures the runs ledger row, and pins active versions once per run;
log_api_call/log_image_generation stamp prompt_version (run pins) +
agent_role (stage map) on every generation_log row. Single-pass keeps
stage 'draft' and stamps content.single_pass explicitly.
Under-300 refactors: Content_Generator's four duplicated span/log blocks
consolidated into execute_stage() (301 -> 224 lines, behavior
identical); get_avg_tokens_for_stages body moved to Cost_Reporter with a
delegating wrapper kept for API compatibility.
Tests: byte-identical fallback renders (single_pass/outline/polish/
editor.system), token filling, key coverage incl. stage-map integrity.
Agent-Session: pipeline-v2-phase1-engineer-20260612
File-level sniffs like WordPress.Files.FileName are evaluated at the file level,
not at code level. They require phpcs:ignore directives to be placed within the
file-level docblock (/** ... */), not in code comments.
Moved all 81 phpcs:ignore pragmas from code comments to the file-level docblocks.
This applies to:
- 79 class files with class-naming convention ignores
- prautoblogger.php with strict-types docblock ignore
- uninstall.php with strict-types docblock ignore
- Added phpcs:ignore for class file naming convention (79 violations)
WordPress.Files.FileName.InvalidClassFileName is a known architectural
issue — class files use short names without prautoblogger- prefix
- Added phpcs:ignore for ->prepare() table interpolation (56 violations)
WordPress.DB.PreparedSQL.InterpolatedNotPrepared: dynamic table names
(prab_generation_logs, prab_cost_logs) require interpolation; ->prepare()
does not support table name placeholders, only column values
- Fixed file-level docblock positioning (2 violations)
prautoblogger.php and uninstall.php: moved declare(strict_types=1) before
file docblock per PHP best practices
Run phpcbf --standard=WordPress-Core across includes/, prautoblogger.php,
and uninstall.php. Fixed 1,362 formatting violations in 75 files:
- Short array syntax: [] → array()
- Line breaking for long function calls
- Consistent indentation
- Function call parameter alignment
No logic changes; purely mechanical formatting.
Writing instructions were framed as a weak "Additional instructions:"
block at the end of the system prompt. Models prioritized competing
format directives in each stage's user prompt and ignored the style
guide. Now: system prompt wraps instructions as "MANDATORY STYLE GUIDE"
with override language, every stage user prompt reinforces compliance,
polish stage explicitly preserves lists and links, and chief editor
revision preserves structural formatting.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
GLM 4.7 Flash was timing out at 120s on large analysis prompts.
Parse failure errors now log the raw LLM response at error level
(not debug) so the root cause is always visible.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Two new textareas in Content settings let admins provide custom
instructions for the analysis and chief editor LLM stages, matching
the Writing Instructions field added in v0.4.4. All three pipeline
stages now have user-controllable system prompt customization.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Models like GPT 5.1 Nano ignore response_format: json_object and wrap
their JSON in ```json fences or add preamble text. New Json_Extractor
utility tries: (1) direct decode, (2) strip markdown fences, (3) find
outermost { ... } substring. Applied to analyzer and chief editor
parsers. Raw response logged on parse failure for debugging.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
WordPress plugin that auto-researches topics from Reddit, generates
SEO-optimized articles via OpenRouter LLMs, runs them through a chief
editor agent, and publishes on a daily schedule.
Includes:
- Full generation pipeline (collect → analyze → score → generate → review → publish)
- Cost tracking and budget enforcement
- Admin UI: settings, review queue, metrics, activity log
- CI/CD pipeline (GitHub Actions → WordPress REST deploy)
- Unit test suite (PHPUnit + Brain Monkey)
- Structured logging with configurable verbosity
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
