Repository navigation
chore(release): v4.12.1 - #46
Merged
Merged
Conversation
The "Current Sprint" pointer and section described sprint-11-05-2025 as active although it closed in November 2025, so every session loaded stale status as current context.
…ost tools Example code, routing tables and tool defaults used claude-3-opus, the deprecated claude-*-4-20250514 snapshots and two IDs that never existed (claude-haiku-4-20250514, claude-haiku-3-5-20241022), so copied code failed on its first request. Move API code to claude-sonnet-5 / claude-opus-5 / claude-haiku-4-5, add the missing max_tokens to the agent-designer example, and fix context windows (Sonnet 5 and Opus 5 are 1M). The price tables in cost_estimator.py, token_counter.py, agent_evaluator.py, agent_orchestrator.py, prompt_optimizer.py and llm-pricing-guide.md now use current per-MTok prices: Fable 5.1 $10/$50, Opus 5.5 $4/$20, Opus 5 $5/$25, Sonnet 5 $2/$10, Haiku 4.5 $1/$5; cache reads at 0.1x input (Opus 5.5 $0.20, Fable 5.1 $0.25). The llm_integration_guide chain-of-thought row now points Claude users at adaptive thinking + effort.
…/model aliases Subagents are .claude/agents/*.md files whose markdown body is the system prompt, with a comma-separated tools allowlist; there is no .yaml form and no custom-instructions field. Model fields now use the sonnet/opus/haiku/ inherit aliases so agents follow the current model in each tier.
…ng advice - Chain-of-thought patterns: on Claude, reasoning runs in adaptive thinking set by effort; asking Opus 5.5 to reproduce its reasoning in the answer can be refused (reasoning_extraction). Prompt-level CoT stays for models without native thinking. - Replace blanket CRITICAL:/IMPORTANT: prefixes with stated reasons and targeted emphasis. - Recommend structured outputs (output_config.format) over temperature and prose JSON enforcement; Opus 4.7+ and Sonnet 5 reject non-default temperature. Prose fallback kept for other providers. - Effort mapping: budget_tokens is gone on current models; where thinking is always on, "none" maps to effort low. - Drop a pointer to a model-specific-behaviors.md file that does not exist.
…models Workflow files that still name the Claude 4 snapshots now price at their real rates instead of falling back to the default, the alias comment says which model each alias resolves to, and the visualizer labels Opus 5.5 and Fable 5.1.
Subagents are .claude/agents/<name>.md files: name, description, an optional comma-separated tools allowlist and model alias in the frontmatter, and the Markdown body as the system prompt. Every example used allowed-tools and custom-instructions, which Claude Code ignores for agents, so an agent copied from them got every tool and no prompt. - Convert all agent examples and the agent template to tools: plus a body prompt; drop the non-agent context: key from the template. - Stop claiming Bash(pattern) in tools narrows Bash; point to permissions.deny and hooks for command limits. - Replace the nonexistent /agents/<name> invocation with natural language, @-mention and --agent; document isolation: worktree. - Skill substitutions: drop the custom-instructions frontmatter example and the nonexistent $PROJECT_DIR/$FILE; use the documented variables. - context_analyzer: agent files match .claude/agents/*.md.
- Restore four wiped skill descriptions - claude-code-mastery teaches the real subagent file format - Current Claude model IDs and prices in examples and cost tools - Current-Claude notes in the prompting guides
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
[4.12.1] - 2026-09-27 (current-Claude accuracy pass)
Fixed
descriptionempty onclaude-code-mastery,senior-mobile,senior-data-scientistandsenior-cloud-architect; each description is restored.claude-code-masteryteaches the real subagent format: a Markdown file in.claude/agents/withname,description,toolsandmodelin the frontmatter and the system prompt as the body. The.yamlfiles,allowed-tools:andcustom-instructions:it showed are ignored by Claude Code, so agents built from them got every tool and no prompt.claude-3-opus-20240229,claude-haiku-4-20250514,claude-haiku-3-5-20241022, …). API examples useclaude-sonnet-5,claude-opus-5andclaude-haiku-4-5; agent files use thesonnet/opus/haikualiases. The cost tools are re-priced (Sonnet 5 $2/$10, Haiku 4.5 $1/$5, Opus 5 $5/$25, Opus 5.5 $4/$20, Fable 5.1 $10/$50) and keep the legacy dated IDs at their old prices so older workflow files still cost correctly.Changed
CRITICAL:prefixes and temperature tuning: on current Claude models thinking depth is set witheffort, asking for reasoning in the output can be refused, emphasis over-triggers, and structured outputs replace JSON-by-instruction. The techniques stay for other providers.Checks (local)
--helpand real inputs; hand-checked the Sonnet 5 step at $0.044).build_manifest.py,generate_site.pyandmkdocs buildregenerated only the expected files.cli/) is unchanged — no npm release.