Add skern init --no-instructions explicit opt-out (#104)

[devrimcavusoglu]

Summary

Fixes #104.

• New --no-instructions flag on skern init: writes nothing, never prompts, regardless of TTY state. Combining it with --instructions, --print-instructions, --target, or --tool-forming-loop is a validation error (exit 2), matching the CLI's convention for bad input.
• The non-interactive contract is now documented in --help, docs/reference/commands.md, and the agent-setup guide: when stdin is not a TTY or --json is set, skern never prompts and never blocks on input — both questions resolve to "no". This was already the behavior (resolveInstructionChoices gates on isTerminal(stdin)), so the issue's worry about a future blocking read is covered by making it a stated guarantee plus the explicit flag.

Issue DoD

• Explicit opt-out flag — --no-instructions
• Documented behavior when stdin is not a TTY — commands reference "Interactivity contract", agent-setup guide, init --help

Also in this PR (docs only)

Addresses claim 1 of #101: GitHub Copilot reads project skills from .github/skills/, .claude/skills/, or .agents/skills/ (GitHub docs), so skern's .agents/skills/ value is valid and no path change is needed. The Copilot platform page and the shared-directory concept section now say so with the citation (mirrors what #105 put in AGENTS.md), and note the one-body-per-name limitation tracked by #47/#101. #101 stays open for claim 2.

Test plan

• TestInit_NoInstructions_WritesNothing (AGENTS.md present and untouched, instructions: null), TestInit_NoInstructions_TextMode, TestInit_NoInstructions_ConflictsWithOptIn (all four opt-in flags → ValidationError, nothing written)
• go test ./..., make lint green
• Manual: skern init --no-instructions in a dir with AGENTS.md → exit 0, file untouched, no prompt; --no-instructions --instructions → exit 2

🤖 Generated with Claude Code

[devrimcavusoglu]

Review (adversarial pass against #104, and #101 claim 1)

#104 DoD: flag ✅; documented non-TTY behavior — the documentation was right in spirit but the code had a real hole, now fixed (196d7a8):

Sev	Finding	Resolution
High	isTerminal used ModeCharDevice; /dev/null is a character device, so skern init < /dev/null (installer / cron / docker run without -i) still printed the prompt (EOF → "no", so it never blocked — but "never prompts" was false)	Real isatty check via golang.org/x/term (+ TestIsTerminal_NonTTYInputs)
Medium	Flag-conflict exit 2 happened after .skern/ was created	Flags validated before any filesystem write; test asserts .skern/ absent
Medium	Docs said "pass --instructions to run non-interactively", but on a TTY --instructions still stopped to ask about the tool-forming loop	Any instruction flag now disables both prompts; an unasked question keeps its default. CHANGELOG › Fixed entry
Medium	Tests would pass with the flag body deleted (--json already gated prompts)	isTerminalFn is injectable; new tests simulate a terminal and prove no prompt text / no stdin read under --no-instructions and under every flag, and that the no-flag path really prompts
Low	--instructions=false --no-instructions rejected (compared Changed, not value)	Values compared; explicit false accepted
Low	tests/manual/setup.sh ran init --quiet (would hang on a TTY)	Passes --no-instructions

#101 claim 1: verified against GitHub's page (project skills in .github/skills, .claude/skills, or .agents/skills; personal in ~/.copilot/skills or ~/.agents/skills) — docs statement is accurate, DoD item 1 met via "documented with rationale".

Stacked on #106 (base fix/frontmatter-passthrough) so the CHANGELOG/AGENTS.md edits merge cleanly in order; CI runs once the base becomes main.

[devrimcavusoglu]

Closed automatically by GitHub when the stacked base branch was deleted after #106 merged. Continued as #110 (same branch, rebased onto main).