docs: require model identification trailers for agent commits

<!-- agent -->
Agent-created commits do not currently record the model that produced them,
which makes it harder to trace AI contributions to a particular model version.

Require an `Agent-Model` trailer in `AGENTS.md` and an
`Agent-Reasoning-Effort` trailer when the session exposes its configured effort.
Use runtime metadata rather than global defaults, and explicitly record an
unknown model version instead of guessing unavailable values.

Agent-Model: gpt-6.1-sol
Agent-Reasoning-Effort: high
Assisted-by: GPT 6.1 Sol
Co-authored-by: GPT 6.1 Sol <codex@openai.com>

Author: Byron

AGENTS.md

@@ -98,3 +98,20 @@ Commit messages must stand on their own. Put the information needed to understan
 and review the change in the commit body, even when it also appears in a pull
 request description. PR and issue links may provide additional context, but must
 not substitute for that explanation.
+
+## Agent trailers
+
+Every commit created by an agent must include an `Agent-Model` trailer identifying
+the model and version reported by its session or runtime. Include an
+`Agent-Reasoning-Effort` trailer when the configured reasoning effort is available.
+
+For example:
+
+```text
+Agent-Model: gpt-6.1-sol
+Agent-Reasoning-Effort: high
+```
+
+Do not guess these values or infer them from global defaults. If the model version
+is unavailable, use `Agent-Model: unknown`; omit the reasoning-effort trailer when
+its value is unavailable.