Update changelog for v2.1.218

This commit is contained in:
Mike
2026-07-22 15:24:54 -06:00
parent 37c256bbc4
commit b9895f5556
+21
View File
@@ -4,6 +4,27 @@ Note: Only use **NEW:** for entirely new prompt files, NOT for new additions/sec
### Claude Code System Prompts Changelog
# [2.1.218](https://github.com/Piebald-AI/claude-code-system-prompts/commit/37c256b)
_+39,506 tokens_
- **NEW:** Skill: Artifact PR review (composed publish flow) — Adds a structured `pr_review` publishing path that gathers and validates the target GitHub pull request, treats PR content as untrusted data, composes a briefing through a vetted renderer instead of hand-authored HTML, anchors it to the reviewed head SHA, and safely handles live reviewer decisions, GitHub comments, and native review verdicts.
- **NEW:** Skill: Artifact whiteboard and Skill: Artifact whiteboard when-to-use guidance — Add an interactive Artifact canvas for architecture and planning sketches, permit one proactive offer when visual collaboration would outperform prose, preserve user-authored marks while drawing Claude’s responses alongside them, and reconcile user sends through conflict-safe republishes.
- **REMOVED:** Data: Structured tool output field schema — Removes standalone guidance for structured `tool_use_result` output objects, including the completed Agent/Task result contract.
- **REMOVED:** System Prompt: Scope fidelity — Removes standalone scope-fidelity guidance after merging its ambiguity handling, scope boundaries, completion requirements, and truthful incomplete-work reporting into System Prompt: Delivering work at full scope.
- Agent Prompt: /code-review part 3 extra-high and maximum effort modes, part 6 medium effort mode, and part 7 high effort mode — Add finder-stage fallback guidance for when the Agent tool is unavailable.
- Data: Managed Agents core concepts, endpoint reference, environments and resources, events and steering, and outcomes, plus Agent Prompt: Managed Agents onboarding flow — Add ordered, all-or-nothing session `initial_events` for one-call message or outcome startup; distinguish resource resolution during creation from lazy sandbox provisioning; and clarify that termination can mean successful completion or an unrecoverable error.
- Data: Managed Agents core concepts and Data: Managed Agents endpoint reference — Add agent-level `effort` configuration, document that per-session effort overrides are ignored, make update `version` optional for either optimistic concurrency or unconditional last-write-wins behavior, and spell out replacement, clearing, and override constraints.
- Data: Managed Agents client patterns, core concepts, endpoint reference, events and steering, reference — cURL, and tools and skills — Correct event contracts for `user.interrupt`, `user.tool_confirmation`, and `user.custom_tool_result`; distinguish queued events from immediately processed events; clarify that thinking events carry progress rather than thinking content; and recast `system.message` as persistent appended system context rather than prompt replacement while removing the Opus 4.8-only restriction.
- Data: Managed Agents events and steering and Data: Managed Agents multiagent sessions — Add thread-scoped live previews with ordering guarantees and troubleshooting, clarify thread-relative message direction, reject nested multiagent rosters instead of silently flattening them, and document thread-specific versus session-wide interruption and thread archival behavior.
- Data: Managed Agents overview and Data: Managed Agents tools and skills — Clarify beta-header requirements for raw Files and Skills API calls, change large-output offloading to a 100,000-character threshold across built-in and MCP tools, warn that Console-created credentials default to header-only injection, and make skill versions optional for both Anthropic and custom skills.
- Data: Managed Agents scheduled deployments — Expand accepted startup events, replace the fixed ten-second jitter description with interval-based jitter capped at nine minutes, and distinguish immediate archival when an agent is archived from archival on the next trigger after deletion.
- Data: Managed Agents webhooks — Document the actual signature headers and stable retry ID, separate event time from delivery-attempt time, expand session, thread, environment, vault, and memory-store lifecycle events, and define subscription timing, duplicate delivery, unordered arrival, bounded retries, event loss, and reversible auto-disable conditions.
- Data: Workshop artifact HTML template and Skill: Artifact workshop — Replace per-option page mutation with a validated decision-state island and batched confirmed choices or typed answers, add conflict-safe reconciliation and iterative republishing, and replace finalization with a “Ready to build” state that lets the reader start building or continue iterating.
- Skill: Building LLM-powered applications with Claude — Clarifies that the listed prices are Anthropic first-party API rates that also apply to Microsoft Foundry, while Amazon Bedrock and Vertex AI use separate partner pricing.
- System Prompt: Delivering work at full scope — Absorbs the former scope-fidelity rules, strengthens routine ambiguity handling and scope boundaries, requires finishing and reporting work accurately, and directs disagreements and refusals to remain fair, factual, and free of moralizing.
- Tool Description: Invoke skill — Clarifies that background skills initially return only the agent name, deliver results later through task notifications, and should not be waited on or invoked again while pending.
# [2.1.217](https://github.com/Piebald-AI/claude-code-system-prompts/commit/21a42bc)
_+13,476 tokens_