4.0 KiB

Raw Blame History

nextStep	scriptsDir	outputFile	retryStrategy	reviewLoop
./step-03b-execute-finish.md	../bin/story-automator	{output_folder}/story-automator/orchestration-{epic_id}-{timestamp}.md	../data/retry-fallback-strategy.md	../data/code-review-loop.md

Step 3a: Execute Review Phase

Goal: Run automate (guardrails) and code review loop for the current story. Interaction mode: Deterministic autonomous execution.

Prerequisites

Step 3 completed (create-story and dev-story done)
State document updated with current story progress

Set: scripts="{scriptsDir}"

Story Loop (Continue from Step 3)

C. Automate (Guardrails)

Skip if overrides.skipAutomate

Apply retry/fallback pattern from {retryStrategy}: Non-blocking, but still retry on failure.

# --command required (see Spawn Pattern in step-03)
session=$("$scripts" tmux-wrapper spawn auto {epic} {story_id} \
  --agent "$current_agent" \
  --command "$("$scripts" tmux-wrapper build-cmd auto {story_id} --agent "$current_agent")")
result=$("$scripts" monitor-session "$session" --json --agent "$current_agent")
"$scripts" tmux-wrapper kill "$session"

SUCCESS:

# Update Story Progress: mark automate done
sed -i '' "s/^| ${story_id} |.*$/| ${story_id} | done | done | done | - | - | in-progress |/" "{outputFile}"

Display: [story {N}/{total}] automate -> done → proceed to D

FAILURE → retry up to 3 attempts (non-blocking, so fewer retries), then log warning:

# Update Story Progress: mark automate skipped
sed -i '' "s/^| ${story_id} |.*$/| ${story_id} | done | done | skip | - | - | in-progress |/" "{outputFile}"

Display: [story {N}/{total}] automate -> skip (non-blocking) → proceed to D

D. Code Review Loop

See {reviewLoop} for complete script-based review cycle with v2.3 per-task agent configuration.

MANDATORY log-summary contract (every review cycle):

Run a single grep/regex pass over review output first.
Return only compact fields to parent flow: next_action, confidence, error_class, issues_count, top_issues.
Do not carry full log payloads forward unless escalation requires raw evidence.

review_log=$(echo "$result" | jq -r '.output_file')
review_focus=$(grep -nE "SUCCESS|FAIL|ERROR|CRITICAL|WARN|RETRY|ESCALATE|ISSUE" "$review_log" | head -n 120)
if [ -z "$review_focus" ]; then
  review_focus=$(tail -n 120 "$review_log")
fi

# Compact subprocess-style summary contract for parent flow
review_summary=$("$scripts" orchestrator-helper parse-output "$review_log" review | jq -c '
  {
    next_action: (.next_action // "retry"),
    confidence: (.confidence // 0),
    error_class: (.error_class // "unknown"),
    issues_count: ((.issues // []) | length),
    top_issues: ((.issues // [])[:3])
  }
')

Key points:

Up to 5 cycles using story-automator tmux-wrapper spawn review + story-automator monitor-session
Agent: Uses per-task config from state document (resolve_agent_for_task "review")
Verification: Uses --workflow review --story-key for sprint-status verification
States: completed (verified):
```
# Update Story Progress: mark code-review done
sed -i '' "s/^| ${story_id} |.*$/| ${story_id} | done | done | done | done | - | in-progress |/" "{outputFile}"
```
Display: [story {N}/{total}] review -> done → E | incomplete → count as failed attempt, retry until maxCycles, then CRITICAL escalate (Trigger #8)
Exit loop when sprint-status shows "done"
If review_summary.next_action is ambiguous, ask one clarifying question before escalating.

Auto-Proceed to Finalization

Display: "Code review complete. Proceeding to finalize commits and status checks..."

"$scripts" orchestrator-helper state-update "{outputFile}" \
  --set currentStep=step-03b-execute-finish \
  --set lastUpdated="$(date -u +%Y-%m-%dT%H:%M:%SZ)"
echo "- **[$(date -u +%Y-%m-%dT%H:%M:%SZ)]** Code review complete, proceeding to finalization" >> "{outputFile}"

Then

→ Immediately load and execute {nextStep}

4.0 KiB Raw Blame History