Submit agent os feedback

Skill designgrappler/agent-os/claude/skills/submit-agent-os-feedback

Agent OS is a management layer for complex, multi-agent AI projects. Instead of letting agents get lost in endless conversational loops, it enforces a disciplined workflow: upfront planning, shared context, scoped work tracks, and strict quality gates — so every agent stays focused, coordinated, and aligned from start to finish.

Install

npx -y skills add designgrappler/agent-os --skill submit-agent-os-feedback

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

1 stars1 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its author says it does

Copied from the file, not written here

File a structured issue against the designgrappler/agent-os public mirror. Use when the user says "file feedback", "report an Agent OS issue", "this Agent OS skill is broken", or wants to report a bug or improvement in Agent OS.

SKILL.md

4.8 KB, as published. Nobody here has run it

Submit Agent OS Feedback

Files a structured GitHub issue against designgrappler/agent-os using the user's existing gh auth session.

Trigger

When the user runs /submit-agent-os-feedback or says:

"file feedback"
"report an Agent OS issue"
"this Agent OS skill is broken"

Phase 1 — Read version header

Read release-version from the canonical manifest:

Primary (URL): Fetch https://raw.githubusercontent.com/designgrappler/agent-os/main/skills-manifest.json and parse .release-version.
Fallback (local clone): If the fetch fails, read ~/Developer/agent-os-private/skills-manifest.json and parse .release-version.
If both fail: Stop. Display: Could not resolve Agent OS version — URL fetch failed and local clone not found at ~/Developer/agent-os-private/skills-manifest.json. Resolve access and re-run /submit-agent-os-feedback.

Display to the user before any prompt:

Filing feedback against Agent OS <release-version>

Hard constraint: Always include the version header (<release-version>) at the top of the issue body — this is the single load-bearing requirement.

Phase 2 — Wizard

Ask three sequential questions. Do not proceed to the next question until the current one is answered.

"What happened? (one-paragraph description)" → captured as WHAT_HAPPENED
"Which skill or agent does this concern? (e.g. update-agent-os, architect, or general)" → captured as SUBJECT
"Optional context — recent commands, error messages, or transcript snippets (paste or skip)" → captured as CONTEXT (may be empty; "skip" or empty reply = CONTEXT is (none provided))

Phase 3 — Secret detection + Preview

Before rendering any preview, scan WHAT_HAPPENED and CONTEXT for secret-shaped strings. Check each pattern:

Pattern	Description
`[A-Z][A-Z0-9_]{3,}=\S+`	Env-var assignment (e.g. `SECRET_KEY=abc123`)
`Bearer\s+\S+`	Authorization header token
`sk-[A-Za-z0-9]{16,}`	OpenAI-style secret key
`ghp_[A-Za-z0-9]{16,}`	GitHub personal access token
`xox[baprs]-[A-Za-z0-9-]{10,}`	Slack token
`AKIA[0-9A-Z]{12,}`	AWS access key
`[A-Za-z0-9_\-]{40,}` adjacent to `key`, `token`, `secret`, or `password`	Generic long credential

On any match: HARD BLOCK.

Display the matched span(s) clearly.
Instruct the user to remove the string and re-run /submit-agent-os-feedback.
Do NOT render a preview.
Do NOT offer "submit anyway."
There is no override path. The user must remove the matching string before submission can proceed.

If no match, build the issue body using this fixed template:

**Agent OS version:** <release-version>
**Subject:** <SUBJECT>
**What happened:**
<WHAT_HAPPENED>

**Context:**
<CONTEXT or "(none provided)">

---
*Filed via /submit-agent-os-feedback from <project-name>.*

Where <project-name> is the last path component of $CLAUDE_PROJECT_DIR (or the current working directory if unavailable).

Title: [feedback] <first 60 chars of WHAT_HAPPENED>

Display the full preview to the user with the title and body. Ask:

Submit, edit, or cancel?

Submit → proceed to Phase 4.
Edit → return to Phase 2; allow the user to revise any field.
Cancel → stop. No issue is created.

Phase 4 — Submit

Never submit without explicit user "submit" confirmation in Phase 3.

Defensive label check: Run:

gh label list --repo designgrappler/agent-os

Parse the output. If feedback is not in the list, stop:

The "feedback" label does not exist on designgrappler/agent-os. Ask the repo owner to create it, then re-run /submit-agent-os-feedback.

Submit: Run:

gh issue create --repo designgrappler/agent-os --title "<title>" --body "<body>" --label feedback

Capture the resulting issue URL. Display it to the user:

Issue filed: <URL>

On error (label missing, auth failure, network): Surface the gh stderr verbatim. Stop. Do not retry without explicit user confirmation.

Hard Constraints

Never submit without explicit user "submit" confirmation in Phase 3.
Hard block on secret-shaped strings — no override path. User must remove the matching string before submission can proceed.
Always include the version header (<release-version>) at the top of the issue body — this is the single load-bearing requirement.
Always invoke gh label list --repo designgrappler/agent-os before submitting and surface a clear remediation if feedback is absent.