Files
prautoblogger/includes/core/class-content-generator.php
peptiderepo 2bdc763dc0 fix: empty-content guard + finish_reason=length handling + reasoning budget (v0.18.1)
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
2026-06-11 13:15:58 +00:00

295 lines
9.4 KiB
PHP

<?php
declare(strict_types=1);
/**
* Writer agent — generates blog post content from article ideas.
*
* Supports: single_pass (one LLM call) or multi_step (outline → draft → polish).
* Opik instrumentation: spans per LLM call when feature-flag enabled.
* v0.18.0: prompt copy renders from the versioned prompt registry (with
* byte-identical in-code fallback); every log row carries the run-pinned
* prompt_version. The four stage methods share one execute_stage() helper —
* span lifecycle, LLM dispatch, and cost logging are identical across stages.
*
* Triggered by: PRAutoBlogger_Article_Worker.
* Dependencies: LLM_Provider_Interface, Cost_Tracker, Content_Prompts.
*/
class PRAutoBlogger_Content_Generator {
private PRAutoBlogger_LLM_Provider_Interface $llm;
private PRAutoBlogger_Cost_Tracker $cost_tracker;
/** @var string|null Run UUID for stage-state checkpointing (null = off). */
private ?string $run_id = null;
/** @var string|null Stage item key scoping this article within the run. */
private ?string $item_key = null;
public function __construct(
PRAutoBlogger_LLM_Provider_Interface $llm,
PRAutoBlogger_Cost_Tracker $cost_tracker
) {
$this->llm = $llm;
$this->cost_tracker = $cost_tracker;
}
/**
* Enable per-stage state checkpointing for one article item (v0.18.0).
*
* When set, every LLM stage (outline/draft/polish or the single-pass
* draft) records pending->running->done state and snapshots its output;
* a re-entered done stage returns the snapshot WITHOUT a new LLM call
* (never re-charged). When unset (default), behavior is exactly
* pre-v0.18.0 — used by callers outside a run (e.g. eval mode).
*
* @param string|null $run_id Run UUID, or null to disable.
* @param string|null $item_key Item key from Run_Stage_State::item_key_for_idea().
* @return $this
*/
public function set_run_item( ?string $run_id, ?string $item_key ): self {
$this->run_id = $run_id;
$this->item_key = $item_key;
return $this;
}
/**
* Generate a blog post from an article idea.
*
* @param PRAutoBlogger_Article_Idea $idea The scored idea to generate content for.
* @param bool $eval_mode True to suppress publish/image side effects.
* @return string Generated HTML content.
*/
public function generate( PRAutoBlogger_Article_Idea $idea, bool $eval_mode = false ): string {
// In eval mode, suppress publishing and image generation side effects.
define( 'PRAUTOBLOGGER_EVAL_MODE', $eval_mode );
$mode = get_option( 'prautoblogger_writing_pipeline', 'multi_step' );
$request = new PRAutoBlogger_Content_Request(
$idea,
$mode,
get_option( 'prautoblogger_tone', 'informational' ),
absint( get_option( 'prautoblogger_min_word_count', 800 ) ),
absint( get_option( 'prautoblogger_max_word_count', 2000 ) ),
get_option( 'prautoblogger_niche_description', '' ),
json_decode( get_option( 'prautoblogger_topic_exclusions', '[]' ), true ) ?? array(),
get_option( 'prautoblogger_writing_instructions', '' )
);
return 'single_pass' === $mode
? $this->generate_single_pass( $request )
: $this->generate_multi_step( $request );
}
/**
* Single-pass: one LLM call produces complete article (Economy tier).
*
* The log row's stage stays 'draft' (historical contract: a draft row
* without an outline row marks a single-pass run) but the prompt key
* is passed explicitly — the stage map would resolve 'draft' to
* 'content.draft'.
*
* @param PRAutoBlogger_Content_Request $request Generation request.
* @return string Generated HTML content.
*/
private function generate_single_pass( PRAutoBlogger_Content_Request $request ): string {
$model = get_option( 'prautoblogger_writing_model', PRAUTOBLOGGER_DEFAULT_WRITING_MODEL );
return $this->execute_stage(
'draft',
'single_pass_generation',
PRAutoBlogger_Content_Prompts::build_single_pass( $request ),
$request,
$model,
array(
'temperature' => 0.7,
'max_tokens' => 4000,
),
'content.single_pass'
);
}
/**
* Multi-step: outline → draft → polish.
*
* @param PRAutoBlogger_Content_Request $request Generation request.
* @return string Generated HTML content.
*/
private function generate_multi_step( PRAutoBlogger_Content_Request $request ): string {
$model = get_option( 'prautoblogger_writing_model', PRAUTOBLOGGER_DEFAULT_WRITING_MODEL );
$outline = $this->execute_stage(
'outline',
'outline_generation',
PRAutoBlogger_Content_Prompts::build_outline( $request ),
$request,
$model,
array(
'temperature' => 0.5,
'max_tokens' => 1500,
)
);
$draft = $this->execute_stage(
'draft',
'draft_generation',
PRAutoBlogger_Content_Prompts::build_draft( $request, $outline ),
$request,
$model,
array(
'temperature' => 0.7,
'max_tokens' => 4000,
)
);
return $this->execute_stage(
'polish',
'polish_generation',
PRAutoBlogger_Content_Prompts::build_polish( $draft ),
$request,
$model,
array(
'temperature' => 0.4,
'max_tokens' => 4000,
)
);
}
/**
* Run one writing stage: Opik span → LLM call → span close → cost log.
*
* Consolidates the per-stage boilerplate that was previously duplicated
* across the four stage methods (v0.18.0). The system prompt is rebuilt
* per stage exactly as before.
*
* Side effects: one OpenRouter API call, one Opik span (when tracing),
* one generation_log row.
*
* @param string $stage Log stage value ('outline'|'draft'|'polish').
* @param string $span_name Opik span name.
* @param string $user_prompt Rendered user prompt for the stage.
* @param PRAutoBlogger_Content_Request $request Generation request (system prompt input).
* @param string $model Model identifier.
* @param array<string, mixed> $options LLM options (temperature, max_tokens).
* @param string|null $prompt_key Explicit registry key for prompt_version
* stamping (null = derive from stage).
* @return string Stage output content.
*/
private function execute_stage(
string $stage,
string $span_name,
string $user_prompt,
PRAutoBlogger_Content_Request $request,
string $model,
array $options,
?string $prompt_key = null
): string {
// Idempotent resume: a done stage returns its snapshot, no LLM call.
if ( null !== $this->run_id && null !== $this->item_key ) {
$cached = PRAutoBlogger_Run_Stage_State::get_output( $this->run_id, $stage, '', $this->item_key );
if ( null !== $cached ) {
PRAutoBlogger_Logger::instance()->info(
sprintf( 'Reusing completed %s stage output (idempotent resume).', $stage ),
'pipeline'
);
return $cached;
}
PRAutoBlogger_Run_Stage_State::start( $this->run_id, $stage, '', $this->item_key );
}
// v0.18.1: stage metadata for the provider's empty-completion
// guard — warning logs and failed-attempt audit rows carry the
// real stage + registry key (never sent upstream).
$options['stage'] = $stage;
$options['prompt_key'] = $prompt_key;
$ctx = PRAutoBlogger_Opik_Trace_Context::current();
$span_id = $ctx->start_span(
array(
'name' => $span_name,
'type' => 'llm',
'model' => $model,
'provider' => 'openrouter',
)
);
$response = $this->llm->send_chat_completion(
array(
array(
'role' => 'system',
'content' => PRAutoBlogger_Content_Prompts::build_system( $request ),
),
array(
'role' => 'user',
'content' => $user_prompt,
),
),
$model,
$options
);
// v0.18.1 writer-path guard (belt over the provider seam, which
// already throws for OpenRouter): no empty article may ever leave
// a writing stage — it would flow to the editor, be rejected as
// "draft text is missing", and land as an empty draft post.
if ( '' === trim( (string) $response['content'] ) ) {
throw new \RuntimeException(
sprintf(
/* translators: 1: writing stage name, 2: model identifier. */
__( 'Writing stage "%1$s" produced empty content (model %2$s) — failing the stage instead of passing an empty article downstream.', 'prautoblogger' ),
$stage,
(string) $response['model']
)
);
}
$ctx->end_span(
$span_id,
array(
'usage' => array(
'prompt_tokens' => $response['prompt_tokens'],
'completion_tokens' => $response['completion_tokens'],
'total_tokens' => $response['prompt_tokens'] + $response['completion_tokens'],
),
)
);
$this->cost_tracker->log_api_call(
null,
$stage,
$this->llm->get_provider_name(),
$response['model'],
$response['prompt_tokens'],
$response['completion_tokens'],
'success',
'',
null,
$prompt_key
);
if ( null !== $this->run_id && null !== $this->item_key ) {
PRAutoBlogger_Run_Stage_State::done(
$this->run_id,
$stage,
'',
$this->item_key,
(string) $response['content'],
$this->llm->estimate_cost( $response['model'], (int) $response['prompt_tokens'], (int) $response['completion_tokens'] )
);
}
return $response['content'];
}
/**
* Generate content in eval mode (no publishing, no image generation).
*
* @param PRAutoBlogger_Article_Idea $idea The scored idea to generate content for.
* @return string Generated HTML content.
*/
public function generate_eval( PRAutoBlogger_Article_Idea $idea ): string {
return $this->generate( $idea, true );
}
}