fix: create-release noop — semver step outputs unavailable during prompt interpolation

[Copilot]

Pre-agent steps: are compiled into the agent job, but prompt interpolation happens in the activation job (which runs first). Any ${{ steps.<id>.outputs.* }} reference in the prompt body is therefore always empty — the step hasn't run yet in that job context.

create-release.md computed next_version / previous_tag in a semver pre-step and used both in the prompt. The agent received a prompt with literal empty strings and correctly called noop.

Changes

• Remove semver pre-step — its outputs can never reach the prompt; the step was dead code from the prompt's perspective
• Rewrite task body — agent now discovers the previous tag via git describe --tags --abbrev=0 and computes the bump itself, guided by ${{ github.event.inputs.bump }} (a workflow dispatch input, properly interpolated at activation time)
• Recompile lock file — GH_AW_STEPS_SEMVER_OUTPUTS_* env vars removed from activation job steps

Before (prompt received by agent):

You are preparing release **v**
(a **patch** bump from ``).
…
Call `noop` if … or if a tag `v` already exists.

After: agent runs git describe --tags itself and computes the version before calling create_release.

[github-actions]

🧠 Matt Pocock Skills Reviewer has completed the skills-based review. ✅

[github-actions]

Skills-Based Review 🧠

Applied /diagnosing-bugs — the root cause is correctly identified and the fix is sound. Two suggestions to harden the agent's version computation.

📋 Key Themes & Highlights

Key Themes

• Root cause correctly addressed ✅ — The pre-step outputs were genuinely unreachable at prompt-interpolation time (activation job vs. agent job). Removing them and moving the discovery into the agent prompt is the right fix.
• Determinism risk — The new Step 1 asks the LLM to compute semver mentally from prose rules. The removed pre-step did this deterministically in shell; the same shell snippet should be provided as an executable block in the prompt so the agent runs it rather than reasons about it.
• Ambiguous noop guard — "if the computed tag already exists" is vaguer than the old v${{ steps.semver.outputs.next_version }}. Worth tightening the wording to reference the variable name the agent is expected to carry forward from Step 1.

Positive Highlights

• ✅ Clear PR description with a concrete before/after — easy to understand what was broken and why
• ✅ Lock file regenerated consistently with the source .md changes
• ✅ ${{ github.event.inputs.bump }} is correctly identified as safe to interpolate (workflow dispatch input, available at activation time)
• ✅ Step numbering updated throughout — no orphaned references

🧠 Reviewed using Matt Pocock's skills by Matt Pocock Skills Reviewer · sonnet46 33.8 AIC · ⌖ 6.06 AIC · ⊞ 6.3K
_{Comment /matt to run again}

Comments that could not be inline-anchored

.github/workflows/create-release.md:92

[/diagnosing-bugs] Relying on the LLM to perform semver arithmetic is a latent correctness risk — e.g., the agent could misparse v0.10.0 as v0.1.0 or mishandle pre-release suffixes.

<details>
<summary>💡 Suggestion: give the agent a shell snippet to run instead of computing mentally</summary>

The removed pre-step's logic was correct; the only problem was where it ran. Provide a copy of that shell logic as an executable block in Step 1 so the agent runs it and reads the output rather t…

.github/workflows/create-release.md:151

[/diagnosing-bugs] The noop guard now says "if the computed tag already exists" — but if the agent makes a computation error in Step 1, it may check for the wrong tag and skip a duplicate-release guard. Tying the guard to a shell-computed value (see Step 1 suggestion) makes it unambiguous.

<details>
<summary>💡 Suggestion</summary>

Be explicit: store the computed version in a variable at Step 1 and reference it by name in the guard:

Call `noop` if the commit list is empty or if tag…

</details>