fix: append < /dev/null to altimate-code run invocations (stdin-wedge workaround)

[sahrizvi]

Summary

The altimate-code skill instructs invocations like:

altimate-code run "<task>" --yolo --output /tmp/altimate-result.md --dir "$(pwd)"

When the host (Claude Code's Bash tool, a CI runner, subprocess.run(..., stdin=None), a plugin spawn without stdio: "ignore") inherits stdin into the child without closing it, altimate-code run hangs at 0% CPU forever. Bun.stdin.text() waits for an EOF that never arrives — no session created, no log, no error, indistinguishable from a deadlock.

Full repro and root cause: AltimateAI/altimate-code#934.

This PR adds < /dev/null to every documented altimate-code run invocation in skills/altimate-code/SKILL.md as a transitional workaround that closes stdin from the caller side. Five invocations updated: minimal, recommended, and the three example patterns (expensive-queries, lineage, table-profile).

Why this is a workaround, not the final fix

The proper fix is at the source — AltimateAI/altimate-code#935 tightens the stdin-read guard in run.ts so the wedge cannot trigger regardless of caller-side stdin handling. That PR is out for review; once it ships in a tagged altimate-code release, the < /dev/null redirect here becomes redundant.

The cleanup is tracked as a separate issue with explicit action items: #11 — "Remove < /dev/null workaround once min altimate-code version includes stdin-wedge fix." When #935 ships, the followup PR will (a) replace each < /dev/null with a min-version prereq check in the existing "Prerequisite Check" section, and (b) remove the HTML comment added by this PR.

What this PR contains

• 5 invocation patterns updated: < /dev/null appended (multi-line continuations get the redirect on a continuation line; single-line invocations get it inline).
• 1 HTML comment placed next to the first occurrence, invisible in the rendered marketplace view, pointing at the tracking issue and #935. Visible to anyone who opens the raw markdown — the audience that matters when it's time to remove the workaround.

What this PR does NOT contain

• No changes to other skills' SKILL.md files (only altimate-code calls altimate-code run as a subprocess).
• No min-version prereq check yet — that lands in the followup PR once a release-with-#935 ships and we have a concrete version number.
• No code changes — this is a SKILL.md change. SKILL.md is the instruction set the LLM reads when the skill fires, so behavioral correctness depends on this content matching the runtime; the diff is small but load-bearing.

Test plan

• Local SKILL.md (with altimate-code-preview swap + this same < /dev/null pattern) has been in use against real workloads since 2026-06-04, no regressions observed.
• Reviewer: from a non-TTY parent process, invoke altimate-code run "say hi" --yolo < /dev/null against the current unpatched altimate-code 0.8.7 binary — confirm it returns in ~1s instead of hanging.
• Reviewer: confirm the < /dev/null doesn't break the pipe-only path (it doesn't — pipe-only invocations don't pass a positional message, and this PR only changes invocations that do pass a positional).

Links

• Bug repro: altimate-code#934
• Source-side fix: altimate-code#935
• Cleanup tracking: Remove < /dev/null workaround once min altimate-code version includes stdin-wedge fix #11
• Catalog entry: plugin-skill-experiments/03-issues-and-fixes.md Issue docs: add model info and benchmark note to Snowflake section #3

[suryaiyer95]

i think should not be part if it, If the goal is to release 935, let's skip this

[sahrizvi]

Okay