Convert turn-halting prompt hooks to a non-blocking command hook; release v0.7.1

[samtalki]

The plugin's three hooks were all type: prompt, so each matching tool call (eval, Write/Edit) spent a separate decision-model call and ended the turn with "stopped continuation", even for scalar results. The prompts were also written as instructions to the main agent rather than the allow/deny criteria a prompt hook expects, so the contract was malformed on top of being wasteful.

What changed

• Removed the two prompt hooks that only restated skill guidance (display code before eval; expand results for color plots). The always-on julia-evaluation/julia-plot skills already carry that guidance.
• Rewrote the post-edit revise hook as a non-blocking type: command hook (claude-plugin/hooks/revise-nudge.sh). On a Write/Edit of a .jl file it prints {"hookSpecificOutput":{"hookEventName":"PostToolUse","additionalContext":"…call revise…"}} and nothing otherwise. PostToolUse cannot block, so the nudge never halts the turn; non-.jl edits are a clean no-op (exit 0). No jq/python dependency.
• Updated test/test_plugin_validation.sh: the old PreToolUse grep would now fail, so it checks hooks.json is valid JSON, has a PostToolUse hook, references revise-nudge.sh, and that the script exists (26 checks, 0 failed).
• Fixed stale docs: README advertised an auto-triggering julia-language skill that no longer exists (now julia-plot); reworded every "automatically calls revise" hook overclaim across README/CLAUDE.md/plugin README to describe the non-blocking reminder.
• Bumped 0.7.0 → 0.7.1 (Project.toml, plugin.json, marketplace.json ×2, both SKILL.md).

Verification

• revise-nudge.sh emits valid additionalContext JSON for .jl, nothing for .md, exit 0 both.
• bash test/test_plugin_validation.sh → 26 passed, 0 failed, version 0.7.1.
• julia --project=. -e "using Pkg; Pkg.test()" → 339/339 pass.

🤖 Generated with Claude Code