fix(amazonq): compaction nudge no longer reads "almost full (300%)"

[laileni-aws]

Problem

COMPACTION_BODY is rendered from measured utilization:

const body = COMPACTION_BODY(Math.round((characterCount / maxOverallCharacters) * 100))

characterCount is currentRequestCount = the current turn's input plus the entire serialized history (message content, tool_use and tool_result blocks). That ratio is not bounded by 1. A long agentic session with many serialized tool specs can sit well above the budget, and users have reported readings of 200-300%.

The copy interpolated that number unconditionally:

The context window is almost full (300%) and exceeding it will clear your history.

At 300% that is self-contradictory. In practice it misleads twice over: users read it as the percentage being miscalculated rather than as a genuinely over-budget session, and "exceeding it" is wrong tense once the budget is already exceeded.

This surfaced on a customer escalation where the over-100% readings were initially triaged as a display/denominator bug. They were not — the numbers were real. The message was what made them look wrong.

Change

Branch the wording at 100%, and keep the real percentage visible:

• < 100% — unchanged: "The context window is almost full (79%) and exceeding it will clear your history."
• >= 100% — "The context window is full (300% of the limit) and continuing will clear your history."

The percentage is deliberately not capped at 100. An over-budget session should stay diagnosable; capping it would have hidden the very signal that identified this case.

Also renames the parameter threshold -> percentUsed. It is measured utilization, not a threshold, and the old name actively invited the misreading that the denominator was the compaction trigger point.

Scope

Deliberately narrow. No change to the budget model, the compaction trigger, or tool-spec serialization. tokenLimitsCalculator.ts already consolidated the previously-conflicting budget constants, so this is the remaining piece.

Compatibility

• No behaviour change below 100%.
• One production call site (agenticChatController.ts), positional, number in / string out - unaffected by the rename.
• No existing test or snapshot asserted the old copy.

Testing

Adds constants.test.ts covering the wording below, at, and above 100%, that the real percentage is preserved rather than capped, and that the compaction offer is always present.

I was not able to run npm run test:unit in my environment (no command runner available in the sandbox), so please let CI validate this before review. Static checks that did pass: the pre-commit prettier/pretty-quick hook parsed and accepted both files, and commitlint accepted the commit. This is a copy + test change with no control-flow or type surface change.

---

🤖 Drafted by DAE-ProdEng-Agent - AI-generated, please verify before merging. Opened as a draft with no reviewers; a human will publish and assign.

[codecov-commenter]

⚠️ Please install the to ensure uploads and comments are reliably processed by Codecov.

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!