fix(distillery-sync): reframe prompt imperatively to prevent empty-output runs

[norrietaylor]

Closes #315.

Root cause

distillery-sync failed every scheduled/push run with "No Safe Outputs Generated", and the store + status issue went stale. Not connectivity — the Distillery MCP connected and registered tools. The prompt opened third-person ("This agentic workflow keeps the Distillery knowledge store current… It runs when…"), and the pinned Haiku model read it as a spec handed to it for review:

"I dont see an explicit task request in your message. Are you asking me to: 1. Implement this workflow … Please clarify."

It then ended the turn (num_turns=1, zero tool calls), never calling distillery_* or noop → empty output → failure. Evidence in #315.

Fix (prompt-only; model stays pinned)

• Open with an imperative, second-person directive: this prompt is the task, perform it now.
• Headless guardrail: non-interactive run, never ask for clarification, act on the documented deterministic defaults.
• Mandatory terminal action: always upsert the status issue (every run, including a no-op that records "no changes" + advances the cursor); noop only if genuinely unable to act. Never end a turn with no tool calls and no safe output.

Model remains claude-haiku-4-5; only the embedded prompt text changes in the recompiled lock (verified: no other lock touched, safe-outputs config and engine unchanged).

Verify

• gh aw compile distillery-sync → 0 errors/0 warnings.
• Lock diff confined to the prompt heredoc; claude-haiku-4-5 still pinned.

After merge: retag v0.3.0 to deploy to consumers (same flow as #309).

🤖 Generated with Claude Code

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Pro Plus

Run ID: 56b0b698-0c8f-43cd-ae94-0594dd9378cc

📥 Commits

Reviewing files that changed from the base of the PR and between a496231 and 7086545.

📒 Files selected for processing (2)

• .github/workflows/distillery-sync.lock.yml
• .github/workflows/distillery-sync.md

🚧 Files skipped from review as they are similar to previous changes (1)

• .github/workflows/distillery-sync.md

---

📝 Walkthrough

Summary by CodeRabbit

• Chores

  • Regenerated Distillery sync workflow configuration and lock artifacts with updated metadata and prompt wrapper delimiters.

• Documentation

  • Updated the Distillery-sync agent instructions to run autonomously and non-interactively, avoid clarification, and always terminate with a “safe output” (using the normal persistent status upsert path, with noop only when truly un-actionable).

Walkthrough

Adds an explicit autonomous/non-interactive directive block to the Distillery-sync agent prompt in both the source instructions markdown and the compiled lockfile. The new text forbids clarification requests, enforces deterministic defaults, and mandates that every agent run terminates with a safe output (status-issue upsert or noop). Here-document delimiter IDs and the lockfile metadata hash are regenerated accordingly.

Changes

Distillery-sync Agent Prompt Hardening

Layer / File(s)	Summary
Source instructions: autonomous-agent directive block .github/workflows/distillery-sync.md	Adds preface notes about the noop safe-output contract, then replaces a single descriptive paragraph with a prescriptive block that mandates non-interactive execution, forbids clarification requests, enforces deterministic defaults, and requires every run to terminate with a safe output (status-issue upsert, or noop only if action is impossible).
Compiled lockfile: regenerated prompt and here-doc delimiters .github/workflows/distillery-sync.lock.yml	Regenerates the lockfile with a new metadata hash, rebuilt prompt-construction here-document with updated delimiter IDs, an expanded system prompt section mirroring the source instruction changes, and refreshed here-doc boundary markers for Safe Outputs config and MCP gateway config blocks. JSON policy contents for configs remain unchanged.

Estimated code review effort

🎯 2 (Simple) | ⏱️ ~10 minutes

Possibly related PRs

• norrietaylor/spectacles#191: Directly overlaps with the same Distillery-sync safe-output/status-issue upsert termination path being tightened in this PR.
• norrietaylor/spectacles#248: Modifies the same embedded Distillery-sync agent prompt, including the safe-output and noop guidance in the same lockfile.
• norrietaylor/spectacles#291: Regenerates the same prompt/template heredoc boundaries in distillery-sync.lock.yml at the same prompt-generation code level.

Poem

🐇 No more asking "what shall I do?"
The rabbit runs headless, the schedule comes due.
Upsert the issue, or noop if stuck —
Never just sit there and run out of luck!
Autonomous hops, every run complete. 🌟

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title clearly summarizes the main change: reframing the distillery-sync prompt imperatively to prevent empty-output runs.
Description check	✅ Passed	The description thoroughly explains the root cause (third-person prompt caused model to ask for clarification), the fix (imperative reframing with headless guardrails), and verification steps.
Linked Issues check	✅ Passed	The PR directly addresses all objectives from #315: reframing prompt imperatively, adding headless guardrail, mandating terminal safe-output action, and maintaining model pinning.
Out of Scope Changes check	✅ Passed	All changes are scoped to the prompt reframing and safe-outputs configuration. The model remains pinned, no other workflow logic or dependencies were altered.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch fix/distillery-sync-imperative-prompt

---

_{Comment @coderabbitai help to get the list of available commands.}