fix(chat): prevent session transcript/plan prefix collisions

[cursor]

Summary

• Fixes a critical cross-session history mix-up: list_transcript_files matched dated transcript names with bare starts_with(session_key), so session s1 also loaded s10-*.jsonl (same class for schedule-1 vs schedule-10).
• Applies the same boundary rule to list_plan_files and the desktop recent-plan fallback scan.
• Adds regression tests that fail if prefix-sibling keys are accepted again.

Bug and Impact

• Trigger: sessions (or schedule keys) where one key is a string prefix of another, with dated files {key}-YYYY-MM-DD.jsonl present for both.
• Impact: agent chat history / chat_history tool can leak sibling-session messages into the LLM context and return wrong user-facing history; plan listing can surface the wrong session's plans.
• Desktop transcript reload already used {session_key}-; the executor path used by the agent did not.

Root Cause

Filename convention is {session_key}-{date}.jsonl, but matching used name.starts_with(session_key) without requiring the - boundary after the key.

Fix

• Match dated transcripts with format!("{}-", session_key) (keep legacy exact {session_key}.jsonl).
• Match plan stems as exact session_key or {session_key}-….
• Align desktop recent-plan candidate filter similarly.

Task Linkage

• Task ID: TASK-2026-071
• Task folder: tasks/TASK-2026-071-session-transcript-prefix-match/

Injected Specs

• spec/verification-integrity.md
• spec/task-artifact-language.md
• spec/architecture-boundaries.md
• spec/rust-conventions.md
• spec/testing-policy.md
• spec/docs-sync.md (N/A: fail-closed matching only; no docs/env/command surface change)

Validation Evidence

• Commands executed:
  • cargo test -p skilllite-executor --lib rejects_prefix_sibling_session_keys → 2 passed
  • cargo test -p skilllite-executor --lib → 6 passed
  • cargo clippy -p skilllite-executor --all-targets -- -D warnings → clean
  • python3 scripts/validate_tasks.py → 71 task folders checked passed
• Key results:
  • s1 listing excludes s10-*.jsonl; schedule-1 excludes schedule-10-*
  • Legacy s1.jsonl / exact plan stems still included

Regression Scope

• Areas likely affected:
  • Agent transcript history load / chat_history
  • Plan file listing
  • Desktop recent-plan fallback for default
• Explicit non-goals / already covered by other open PRs:
  • fix(evolution): constrain pending skill path operations #89, fix(artifact): reject Windows path escapes in store keys #123–fix: scope evolution disable/explain and migrate memory roots #134 path/workspace/evolution fixes
  • Evolution backlog dedupe_key UNIQUE reuse after executed (follow-up finding, not in this PR)

Docs Sync (EN/ZH)

• Not needed
• Updated EN + ZH docs
• Files:
  • N/A

Review Checklist

• Acceptance criteria in tasks/TASK-2026-071-session-transcript-prefix-match/TASK.md satisfied
• STATUS.md updated with validation evidence
• REVIEW.md includes merge readiness decision
• tasks/board.md status is up to date