AI-334: rename alert_confidence / triage_confidence → incident_likelihood

[gbrewermc]

Summary

Toolkit-side docs rename matching the upstream ai-agent + mcp-server PR (monte-carlo-data/ai-agent#1115). The MCP alert_assessment tool now returns incident_likelihood instead of alert_confidence, and the remediation skill's stale triage_confidence typo is corrected at the same time. Linear ticket: AI-334.

Files touched

• skills/automated-triage/SKILL.md — tool description + workflow steps
• skills/automated-triage/references/triage-stages.md — output schema + troubleshooting-filter prose
• skills/automated-triage/references/triage-example.md — fixture prose + summary-table header
• skills/remediation/SKILL.md — tool-output description + urgency rules

Key decisions

• HIGH/MEDIUM/LOW casing intentionally unchanged. ai-agent attempted title-casing the Literals mid-implementation and reverted: criticality and other axes the LLM sees in the same context are ALL-CAPS, so forcing one axis to title case creates a chronic "don't ALL-CAPS this one specific term" prompt-discipline burden. Toolkit docs follow the same convention — the field name changes, the value casing doesn't. See ai-agent PR #1115's "Key Decisions" section for the full reasoning.
• triage_confidence typo fix in remediation/SKILL.md was scoped into this PR since it's the same concept the rename addresses (and a third-spelling drift that should be cleaned up at the same time).
• No HIGH/MEDIUM filter sentences flipped. The "BOTH incident_likelihood AND alert_impact are MEDIUM or HIGH" phrasing keeps ALL-CAPS — values are unchanged.

Cutover

Toolkit PR is independent of monolith/FE timing. The MCP server side (ai-agent PR #1115) holds merge until the monolith team signals ready; once that ships, this toolkit PR can merge any time. Users on the released mc-agent-toolkit won't see field-name mismatches because the toolkit MCP tool description and the prose docs both flip together.

Verification

• git grep -in "alert_confidence\|triage_confidence" returns zero hits across the repo.
• Bare "confidence" references in the 4 triage-context files now consistently say "incident likelihood" where they describe the score axis.
• No code or test changes; pure docs rename.

🤖 Generated with Claude Code

[linear]

AI-334