[claude-hackernews] Reply draft: AgentRQ Show HN, task-vs-call drift, PreToolUse layer (id=47958608)

[NiveditJain]

Discovery

Found via /show feed sweep + Algolia search claude code loop (past week, by date). The thread is a Show HN by mrtnx for AgentRQ - a supervisor-MCP that orchestrates self-learning worker agents (Claude Code / Gemini CLI) that can create and schedule their own tasks. Commenter chloeeekim opened the thread with: "I've found that fully autonomous loops tend to need a lot of guardrails to stay useful." OP replied to the autonomy and self-learning sub-questions but did not address the guardrail point - the door is still open in the thread for a substantive layer-split answer.

Target thread

• Story: https://news.ycombinator.com/item?id=47958608 (Show HN: Task Manager for AI Agents (MCP, Opensource), 6 points, 4 comments at draft time, 3 days old)
• Parent comment being replied to: https://news.ycombinator.com/item?id=47960424 (chloeeekim, the guardrail comment)
• OP: mrtnx, repo at https://github.com/agentrq, Apache 2.0
• Reply form: rendered (textarea[name=text] present); no [dead] / [flagged] markers; no login wall.

Proposed comment

Disclosure-on-top, one substantive paragraph, one custom-policy snippet (no-shared-force-push), no install command, no policy-name comma list, no dashboard plug, no ~/.failproofai/ callout. Repo URL appears once. ASCII-only punctuation (hyphens, straight quotes, no em/en-dashes, no curly quotes, no unicode arrows). Body word count ~135 words excluding code.

The angle is task-vs-call drift: AgentRQ's supervisor-MCP supervises which tasks the worker picks up, but a PreToolUse hook supervises what each tool call inside the task is allowed to do. Concrete failure: a self-scheduled "consolidate the staging branch" task whose description passes the persona check but resolves to git push --force origin staging at the call site - the supervisor can't see the drift, the PreToolUse hook can. Different layer; the two stack.

The full reply text plus parent excerpt and per-section notes are in drafts/2026-05-03T211924Z.md.

Status

• Status: draft, pending manual post by the user.
• Per CLAUDE.md "Comments via PR (never direct post)" - this PR is the review-and-approval gate. No HN textarea was touched; no submit was clicked.
• After the user posts manually and asks, the comment-permalink gets appended to the HN: line as a follow-up commit.

Duplicate-check results

• drafts/ and comments/ on the current branch: no entry for item?id=47958608.
• Open PRs: scanned gh pr diff for every open PR - no match for item?id=47958608.
• Cross-thread paraphrase guard: this is a task-vs-call drift framing, not a static-vs-runtime gate framing (PR [claude-hackernews] Reply draft: Smithery MCP scan, static-vs-runtime gate (id=47969781) #35, [claude-hackernews] Reply draft: Git Shield Show HN, in-loop vs commit-time gate (id=47972142) #36, [claude-hackernews] Reply draft: Trent Show HN, static-review vs runtime-action layer (id=47962091) #39). The custom-policy snippet (no-shared-force-push against the (main|staging|production) triple) does not appear in any prior draft. The shared-branch force-push deny pattern is fresh on this branch.

Summary by CodeRabbit

• Documentation
  • Added a new draft post discussing AI agent autonomy, guardrails, and policy management with practical examples and insights on execution patterns and task management.

[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

Run ID: ab74aa24-9c85-43a5-9455-fab63e9ac73b

📥 Commits

Reviewing files that changed from the base of the PR and between ebbce06 and 2d882c1.

📒 Files selected for processing (1)

• drafts/2026-05-03T211924Z.md

---

📝 Walkthrough

Walkthrough

This PR adds a single markdown draft file (drafts/2026-05-03T211924Z.md) that replies to a Hacker News comment about AgentRQ, autonomy, and guardrails. The draft contains HN metadata, project background excerpts, the parent question, a detailed reply with a concrete customPolicies.add PreToolUse policy code example, and supporting insights and findings documentation.

Changes

Draft HN Reply: Autonomy & Guardrails

Layer / File(s)	Summary
Metadata & Context drafts/2026-05-03T211924Z.md (lines 1–32)	HN thread and comment links, AgentRQ project description, OP body excerpts, and quoted parent question about tool autonomy and guardrail enforcement.
Reply & Code Example drafts/2026-05-03T211924Z.md (lines 33–56)	Detailed reply paragraph with a concrete customPolicies.add(PreToolUse) code snippet denying force-push commands on shared branches.
Insights & Notes drafts/2026-05-03T211924Z.md (lines 57–73)	Team insights checklist (integration framing, failure-mode terminology, "task-vs-call drift"), formatting constraints (ASCII-only, word count), conformance checks, and thread activity metadata.

Estimated code review effort

🎯 1 (Trivial) | ⏱️ ~2 minutes

Possibly related PRs

• [claude-hackernews] Reconcile drafts/ vs comments/ split; add cron no-op alert #4: Introduces changes to reconcile drafts/ vs. comments/ directory structure, which may conflict with this PR's addition of a new draft file.
• [claude-hackernews] Restore drafts/ vs comments/ split: route writes to drafts/ #6: Restores routing of agent writes and docs to drafts/, aligning with this PR's creation of a new draft.
• [claude-hackernews] Densify cron prompt to mirror working claude-reddit prompt #5: Contains cron prompt changes that instruct the agent to produce markdown drafts with this PR's structure and metadata conventions.

Poem

🐰 A draft post hops through HN threads,
With guardrails woven, policies spread,
PreToolUse blocks the reckless shove,
While autonomy blooms with caution's glove. ✨

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title accurately summarizes the main change: a draft reply to a specific HN comment about AgentRQ, focusing on task-vs-call drift and PreToolUse layer implementation with a concrete policy example.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

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

---

_{Review rate limit: 3/5 reviews remaining, refill in 19 minutes and 38 seconds.}

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