v0.2.2-pre-approval-planning - Pre-Approval Planning Report

Safe Code Agent v0.2.2

Safe Code Agent is a coding-agent skill for safer AI code editing.

This release adds a pre-approval planning layer for risky or unclear coding work.

Highlights

• added Pre-Approval Planning Report to SKILL.md
• added pre-approval defaults to AGENTS.md
• added docs/pre-approval-planning.md
• clarified that approval must include scope, risk, boundaries, and verification plan
• replaced static comparison images with a short Safe Code Agent demo

Note

This is still a prompt/skill design. It helps agents explain risky actions before approval, but it does not replace user review or runtime verification.

v0.2.1-balanced-router - Risk Signal Router for advanced docs

v0.2.1-balanced-router

This release keeps the core Safe Code Agent workflow lightweight while making advanced docs easier to apply.

Added

• Risk Signal Router in README.md, AGENTS.md, and SKILL.md
• docs/advanced/risk-signal-router.md
• Trigger rules for recommending:
  • runtime-enforcement.md
  • prototype-to-production.md
  • structured-hallucination.md

Why

The previous balanced version made advanced docs optional, but users still had to know when to apply them.

This release lets the agent detect risk signals and recommend the relevant advanced doc instead of expecting the user to choose manually.

Examples

• Missing command output for a claimed test/build pass
  -> recommend runtime-enforcement.md

• Prototype code becomes core logic
  -> recommend prototype-to-production.md

• Polished report but weak evidence
  -> recommend structured-hallucination.md

Note

This is still a prompt/skill design, not a runtime enforcement system or benchmark result.

The goal is to keep the core workflow lightweight while making risky situations easier to escalate.

v0.1.0 - Initial release

v0.1.0 - Initial release

Initial public release of Safe Code Agent.

Includes:

• Light / Full mode selection
• patch size guidance
• inspection minimums
• evidence discipline
• failure mode minimums
• verification labels
• large task memory rules
• lightweight AGENTS.md defaults