123cx330yrx/riftcoach-agent/single-match-review

Single Match Review

Objective

Produce a grounded review of exactly one completed match. Treat the selected match row and its deterministic report facts as the only sources of player-specific performance claims.

Workflow

1. Validate Player Summary Schema v1.0 and locate target_match_id exactly once.
2. Isolate the target match instead of turning other match rows into single-game facts.
3. Check short-game and Timeline availability before interpreting the metrics.
4. Use knowledge.search only to explain a metric, limitation, or general training principle.
5. Separate measured facts, cautious interpretations, and bounded training actions.
6. Return the typed output for independent Harness evaluation and publication control.

Evidence Rules

• Preserve all target-match numbers from deterministic inputs exactly.
• Cite retrieved knowledge by source_id when it affects a conclusion.
• Treat a short game as reviewable but insufficient for long-term trend claims.
• Do not treat unavailable Timeline data as zero or infer event timing from Match Detail.
• State insufficient evidence when the target row or retrieved knowledge cannot support a claim.

Forbidden Behavior

• Do not infer the lane opponent, hidden information, live match state, or real-time cooldowns.
• Do not invent patch-meta, rank, matchup, win-rate, build, or rune comparisons.
• Do not call Riot API or tools outside the manifest allowlist.
• Do not use recent aggregate rows as if they were facts from the target match.
• Do not publish a model draft without the configured quality gate.