docs(operations): document subscription-only autonomous-tick path post Agent SDK credit (AISDLC-353)#529
Conversation
AI-SDLC: Reviewer Agents Skipped (Anthropic Budget Exhausted)1/3 review agents failed with HTTP 400 What this means:
Posted by AI-SDLC Review Agents — AISDLC-147 budget circuit breaker (AISDLC-157 broadened gate) |
1cbef6c to
891425c
Compare
Codecov Report✅ All modified and coverable lines are covered by tests. 📢 Thoughts on this report? Let us know! |
891425c to
2f4cfa4
Compare
…t Agent SDK credit (AISDLC-353) Seven documentation additions covering the /ai-sdlc orchestrator-tick subscription-only loop that survives the 2026-06-15 Agent SDK credit change: 1. billing-and-cost-optimization.md §1b: new "subscription-only autonomous loop" section with architecture explanation, trade-off table, operator quickstart, cost-projection table (20 tasks/day), and failure-mode/recovery docs. 2. TL;DR table: added row for the zero-incremental-cost path post-2026-06-15. 3. orchestrator-tick.md: updated description frontmatter to call out this is the subscription-only path, Agent SDK credit NOT drawn, and cross-reference to billing doc §1b. 4. CLAUDE.md "Canonical execution paths" table: added row for autonomous loop via /ai-sdlc orchestrator-tick; annotated --spawner claude row to note it draws the Agent SDK credit pool post-2026-06-15. 5. pipeline-cli/docs/spawner.md: cross-referenced the subscription-only path as the preferred high-throughput alternative in the --spawner claude row. 6. Failure-mode docs: session crash → worktrees survive → AISDLC-242 recoverable-abort detection → operator restarts session + fires command. 7. Task moved from backlog/tasks/ to backlog/completed/. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
2f4cfa4 to
8802a12
Compare
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
AI-SDLC: Reviewer Agents Skipped (Local Attestation Valid)A valid DSSE attestation envelope was found at
CI re-running the same 3 agents would produce the same verdict at the cost Note: Attestation remains AUDIT-ONLY as a merge gate (per AISDLC-140 sub-4). Posted by AI-SDLC Review Agents — AISDLC-147 attestation cost-saver |
Summary
Documents the
/ai-sdlc orchestrator-ticksubscription-only autonomous loop as the preferred zero-incremental-cost dispatch path post-2026-06-15, when the Agent SDK credit change makescli-orchestrator tick --spawner claude(cron/daemon) draw from the $200/mo credit pool.The key insight: because
/ai-sdlc orchestrator-tickruns theAgenttool call inside an interactive Claude Code session turn (not a non-interactive Agent SDK invocation), it draws from the operator's interactive Max-20x quota rather than the Agent SDK credit pool. As long as one Claude Code session stays alive, the loop churns the backlog at zero incremental cost above the subscription fee.Changes
docs/operations/billing-and-cost-optimization.md: new §1b with full architecture explanation, operator quickstart (3 commands), side-by-side comparison table vs--spawner claudecron path, cost-projection table at 20 tasks/day throughput, and failure-mode/recovery docs.ai-sdlc-plugin/commands/orchestrator-tick.md: updated description frontmatter to call out the subscription-only billing path and cross-reference billing doc §1b.CLAUDE.md"Canonical execution paths" table: new row for the autonomous loop; annotated--spawner clauderow to note Agent SDK credit draw post-2026-06-15.pipeline-cli/docs/spawner.md: cross-referenced subscription-only path as preferred high-throughput alternative in the--spawner clauderow.backlog/tasks/tobacklog/completed/.Failure-mode coverage
Documents session crash recovery: worktrees survive, AISDLC-242 recoverable-abort detection, operator restarts session and fires the command again.
Test plan
🤖 Generated with Claude Code