[claude-hackernews] Reply draft: AI CAD Harness Show HN, bounded-mutation in agentic feature-tree edits (id=47977694)

[NiveditJain]

Discovery

Browser-driven sweep across /ask, /show, hn.algolia.com (claude code hooks, agent deleted, claude code permissions, yolo mode claude, ai agent production) and /best — most agent-reliability and security-gateway Show HNs are already covered in this repo's open PR set (40 thread IDs already drafted). Picked "Show HN: AI CAD Harness" by Adam.new (zachdive), which sits in the Show HN of an adjacent product where OP solicits design discussion row of INSTRUCTIONS.md's Thread-fit gate — adam.new is an in-CAD agent harness (their own framing) that edits feature trees agentically via FeatureScript / Python, and OP closes with: "What would you want an in-CAD agent to do that nothing does today?"

• Discovery URLs: /show, /best, hn.algolia.com queries above.
• Thread URL: https://news.ycombinator.com/item?id=47977694
• 94 points, 94 comments, 2 days old.

Why this thread

OP runs an agent harness on someone's CAD model and lists agentic use cases like "merge redundant features and clean up my tree", "rename every feature so the tree is actually readable", "round all internal edges with a 2mm fillet", "parametrize my model". None of these are wrong as commands; they are bulk transforms on a tree where each edit looks local but has non-local consequences (downstream FeatureScript references, parametric invariants, per-feature radii). The reliability shape is identical to coding-agent land: an agent doing a structured-tree edit where "wrong" is invisible until something else breaks. OP explicitly invites design suggestions, which gates a substantive design comment with a brief cross-domain mention of FailProof's per-resource policy pattern.

Proposed reply (132 words)

(disclosure: I work on FailProof AI: https://github.com/exospherehost/failproofai)

The interesting design question for an in-CAD agent isn't generation, it's bounded mutation. "Rename every feature" is locally fine but silently breaks any FeatureScript expression that referenced the old name; "merge redundant features" can erase an invariant a downstream part assumed; even "round all internal edges with a 2mm fillet" overwrites per-feature radii the operator chose. The agent isn't wrong, it's that "wrong" is invisible until the next assembly opens.

Do you let users mark features as load-bearing so the agent has to ask before touching them? In coding-agent land the pattern that worked was per-resource policies the operator defines (a pre-tool-call hook that can deny with a message). The CAD analogue would be a tag the agent must respect during rename and merge passes.

Voice / structure check (per INSTRUCTIONS.md)

• One disclosure line, parens, lowercase disclosure:, GitHub URL.
• Two paragraphs of substantive on-topic CAD content that stand alone without the FailProof reference.
• One FailProof concept named (per-resource policies via a pre-tool-call hook). No policy-name comma list. No install command. No three-scope / dashboard / version-number talk.
• Body length 132 words — in the working-shape band (~110-150), well clear of the flagged-shape band (~220).
• ASCII punctuation only — hyphens, periods, parens, slashes; no em/en dashes, no curly quotes, no unicode arrows.
• Single link (in the disclosure line, no repo URL repeated at the bottom).

Status

drafts/2026-05-04T083214Z.md — draft (pending manual post). Per CLAUDE.md "Comments via PR (never direct post)", Claude does not submit on HN. Merge this PR after posting to mark "I posted it"; if you want, ping back with the comment permalink and I'll append it to the HN: line in a follow-up commit.

Summary by CodeRabbit

• Chores
  • Added a draft document for internal discussion and planning.

[coderabbitai]

📝 Walkthrough

Walkthrough

A new Markdown draft file is created documenting a Show HN reply that proposes a bounded-mutation approach to CAD agent reliability, identifies specific failure modes of bulk edits, and outlines a user-controlled protection mechanism via feature tagging and policy hooks.

Changes

Draft Content Addition

Layer / File(s)

Summary

Draft File drafts/2026-05-04T083214Z.md

New Show HN reply draft containing metadata (status, HN/OP links), a disclosure statement, main reply proposing bounded-mutation reliability framing for CAD agents, cross-domain insight for the FailProof team, and a detailed editorial checklist addressing thread fit, body length constraints, formatting rules, and identified CAD failure modes (FeatureScript expression breakage, invariant loss during merges, per-feature fillet radius overwrites).

Estimated code review effort

🎯 1 (Trivial) | ⏱️ ~3 minutes

Possibly related PRs

• [claude-hackernews] Switch HN automation to drafts-only mode #2: Both PRs involve the drafts/ workflow—this PR adds a new HN draft file while that PR reworks the workflow and schema to make drafts-only and commit drafts/ as the canonical target.
• [claude-hackernews] Restore drafts/ vs comments/ split: route writes to drafts/ #6: This PR adds a draft file under drafts/, directly aligned with that PR's routing of agent writes and documentation to drafts/ as the primary write target.
• [claude-hackernews] Reconcile drafts/ vs comments/ split; add cron no-op alert #4: This PR creates a new drafts/ artifact while that PR modifies the drafts-vs-comments workflow and .gitignore handling for the drafts/ directory.

Poem

🐰 A draft takes shape beneath the moonlight glow,
Bounded mutations steal the show!
Feature tags dance, policies stand guard,
CAD agents bloom less broken, less scarred.
Notes tucked in, ready for review—
The rabbit applauds this clarity true! 🥕

🚥 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 directly and specifically describes the main change: a new Show HN reply draft for a HackerNews post about AI CAD agents, focused on bounded-mutation in feature-tree edits.
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: 4/5 reviews remaining, refill in 12 minutes.}

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

[coderabbitai]

Actionable comments posted: 1

🤖 Prompt for all review comments with AI agents

Verify each finding against the current code and only fix it if needed.

Inline comments:
In `@drafts/2026-05-04T083214Z.md`:
- Line 13: The fenced code block starting at the unlabeled triple backticks
should include a language identifier to satisfy markdown lint rule MD040; update
the fence in the draft (the reply block that contains "(disclosure: I work on
FailProof AI...)" ) from ``` to ```text (or another appropriate language) so the
block is labeled and the lint warning is resolved.

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Pro

Run ID: 44844754-36db-4089-8aa8-89911d7997ab

📥 Commits

Reviewing files that changed from the base of the PR and between ebbce06 and d78bb07.

📒 Files selected for processing (1)

• drafts/2026-05-04T083214Z.md

[coderabbitai]

⚠️ Potential issue | 🟡 Minor | ⚡ Quick win

Add a language identifier to the fenced reply block.

Line 13 triggers MD040; please label the fence (for example text) to keep markdown lint clean.

Suggested fix

-```
+```text
 (disclosure: I work on FailProof AI: https://github.com/exospherehost/failproofai)
 ...

</details>

<!-- suggestion_start -->

<details>
<summary>📝 Committable suggestion</summary>

> ‼️ **IMPORTANT**
> Carefully review the code before committing. Ensure that it accurately replaces the highlighted code, contains no missing lines, and has no issues with indentation. Thoroughly test & benchmark the code to ensure it meets the requirements.

```suggestion

🧰 Tools

🪛 markdownlint-cli2 (0.22.1)

[warning] 13-13: Fenced code blocks should have a language specified

(MD040, fenced-code-language)

🤖 Prompt for AI Agents

Verify each finding against the current code and only fix it if needed.

In `@drafts/2026-05-04T083214Z.md` at line 13, The fenced code block starting at
the unlabeled triple backticks should include a language identifier to satisfy
markdown lint rule MD040; update the fence in the draft (the reply block that
contains "(disclosure: I work on FailProof AI...)" ) from ``` to ```text (or
another appropriate language) so the block is labeled and the lint warning is
resolved.