CHANGELOG.md — Template

Summary

A dated record of what changed in the agent’s operating files (policies / skills / eval) and why. Prevents silent drift, makes changes reversible, and gives every change a clean before/after window. Governed by the weekly policy-change cadence.

How to use

• One CHANGELOG per agent (or per pipeline).
• Add an entry only on the weekly change slot — unless it’s a big/significant or compliance change (those go in immediately, marked [urgent]).
• Each entry: what file, what changed, why, and the evidence (link the decision-log entry).
• After a change, re-run eval.yaml to confirm no regression.

Template

# CHANGELOG — <agent name>
 
## [2026-06-12]  (weekly batch)
### policies.md
- Changed: banned_words += ["miracle", "fixes"]
- Why: 2 posts flagged borderline-claim in week of 06-05
- Evidence: [[Decisions 2026-06#borderline-claims|borderline-claims]]
### skills.md
- Changed: write_hook → add "question-first" pattern
- Why: question hooks beat statement hooks on hook-rate (saves +18%)
- Evidence: [[Decisions 2026-06#hook-ab-test|hook-ab-test]]
### eval.yaml
- Changed: added golden ./golden/peptide_myth_busting.md
- Regression re-run: PASS (8/8 cases)
 
## [2026-06-05]  (initial)
- Established policies.md, skills.md, eval.yaml v0.1

The changelog IS your agent's decision log

It’s the lab notebook applied to the agent system itself. When you wonder “why did output quality shift three weeks ago?”, this is the file that answers it.

Related: Policy-change cadence · policies.template · eval.template