From b013b5a9da41bcccb5d32732008137a5e2f35aa7 Mon Sep 17 00:00:00 2001 From: Mike Date: Fri, 8 May 2026 17:42:25 -0600 Subject: [PATCH] Update changelog for v2.1.136 --- CHANGELOG.md | 10 ++++++++++ 1 file changed, 10 insertions(+) diff --git a/CHANGELOG.md b/CHANGELOG.md index a61c3a0..897790b 100644 --- a/CHANGELOG.md +++ b/CHANGELOG.md @@ -4,6 +4,16 @@ Note: Only use **NEW:** for entirely new prompt files, NOT for new additions/sec ### Claude Code System Prompts Changelog +# [2.1.136](https://github.com/Piebald-AI/claude-code-system-prompts/commit/5db109e) + +_+525 tokens_ + +- **NEW:** System Prompt: Action safety and truthful reporting — Requires confirmation for irreversible or outward-facing actions unless durably authorized, asks agents to inspect targets before deleting or overwriting them, and emphasizes faithful reporting of skipped steps, failed tests, and verified outcomes. +- Agent Prompt: Auto mode rule reviewer — Adds `hard_deny` as a fourth custom-rule category for unconditional security-boundary blocks, and narrows `soft_deny` to destructive or irreversible actions that clear user intent can authorize. +- Agent Prompt: Security monitor for autonomous agent actions (first part) — Splits blocking logic into unconditional hard blocks and user-authorizable soft blocks, updates the default rule, and makes user intent unable to clear hard-block security boundaries. +- Agent Prompt: Security monitor for autonomous agent actions (second part) — Moves data exfiltration into hard-block rules, adds hard-block coverage for safety-check bypasses, and treats agent-guessed external services or download sources as untrusted. +- Tool Description: Edit — Restores the line-number prefix format to a template variable while preserving the guidance to exclude line prefixes from edit strings. + # [2.1.133](https://github.com/Piebald-AI/claude-code-system-prompts/commit/72ca448) _+121 tokens_