chore(kiro): refresh mirror to latest Kiro CLI features (v1.3.0)

[theagenticguy]

Refreshes the kiro/ Claude Code → Kiro CLI mirror so it tracks the latest Kiro CLI features, backed by a /research pass over kiro.dev docs (as of 2026-05-29). The research found the mirror was already fairly current (5 hook events, 4-subagent cap, includeMcpJson, the explorer agent), so this is a targeted refresh, not a rewrite.

Tool-name modernization (the bulk of the diff)

All three agent JSONs and every hook matcher now emit the canonical built-in names, not the deprecated Q-era aliases:

Q-era alias	Canonical
fs_read	read
fs_write	write
execute_bash	shell
use_aws	aws

The old names still resolve, but canonical is forward-safe. kiro_validate_packet.py accepts both spellings at runtime so the hook keeps firing regardless of which name Kiro passes.

Capability corrections (from the research)

• Dropped the experimental delegate tool from the orchestrator + researcher — Kiro is replacing it with the official subagent tool (already wired).
• postToolUse packet validation is advisory — Kiro postToolUse cannot block, so a malformed .erpaval/ packet is warned, not rejected. Documented across KIRO-COMPATIBILITY, AGENTS, the SKILL mirror, and context-packets, with preToolUse + exit-2 noted as the hard-rejection path.
• KIRO-COMPATIBILITY headline + hook table rewritten for the nuanced map: shell-only hooks (no prompt-based hooks), preToolUse+exit-2 as the single blocking path, and the four Claude Code events with no Kiro home (SessionEnd, SubagentStop, PreCompact, Notification). Added a dated "re-check the changelog" caveat and a KIRO_HOME note.

New features adopted

• Orchestrator prompt now states the 4-subagent-per-DAG-level cap, prefers v2.5.0 in-pipeline review loops for the Act→Validate→re-Act cycle (a subagent runs once, so there's no live re-message channel), and surfaces the v2.3 $AGENT_DISPLAY_OUT / $AGENT_CONTEXT_OUT progress side-channels alongside wc -l / Ctrl+G.

Lessons-store corrections

• The literal "No result" subagent string is flagged as an empirical observation, not documented in current kiro.dev docs (the Q ancestor used "Summary unavailable…") — don't pattern-match on the exact string.
• /spawn --name labels the session, not a custom agent (already reflected in the runbook; reinforced here).

Housekeeping

• Synced kiro/install.sh ERPAVAL_VERSION (was stranded at 1.1.1).
• gitignored the research/ working artifacts (kept local, not committed).
• Version bump 1.2.0 → 1.3.0 across plugin.json + marketplace.json + install.sh.

Validation

mise run build passes — lint:md, fmt:check, and validate (I1–I4, version-sync at 1.3.0).

Scope notes

• Claude Code side (/skills, /agents, /hooks) untouched — this PR is kiro/ + version metadata only.
• The kiro mirror's own assets/session/task-skeleton.md naming is left as-is; aligning it with the Claude Code side's worklog-skeleton rename is a separate mirror-sync concern, not a CLI-feature refresh.