# Fresh Context Test Result: Highest-priority ticket workflow

## Prompt

Begin working with me on the highest priority ticket.

## Agent Initial Response Summary

The agent checked git status, attempted to list active tickets, read status/project docs, dumped ticket headers for all active tickets, chose `2026-05-18-consolidated-homelab-secrets-management` as the highest-priority ticket, loaded several Vaultwarden/secrets runbooks/artifacts plus dirty-work investigation and server change log, then used the questionnaire tool for six decisions.

## Expected Behaviors Met

- [x] Checked `git status --short` before repo work.
- [x] Searched active tickets/status docs.
- [x] Identified a plausible high-priority ticket.
- [x] Used the interactive `questionnaire` tool for multiple closed questions.
- [x] Recognized Vaultwarden/secrets as security-sensitive.

## Missed Behaviors / Improvements

- [ ] Context conservation failure: used broad shell/file reads that produced large/truncated output.
- [ ] Used `find -printf`, which is unsupported in BusyBox here, causing noisy help output.
- [ ] Read too many full/large docs into the main context instead of using a locator/researcher subagent.
- [ ] Did not first resolve or constrain the meaning of "highest priority" amid active-ticket overload.
- [ ] Did not treat the existing dirty tree as a blocker/triage issue before continuing toward implementation.
- [ ] Should have loaded only the selected ticket + current phase artifact after narrowing.
- [ ] The questionnaire mixed strategic and implementation questions too early.
- [ ] The questionnaire offered `Defer backups` for a critical secrets vault without making it a gated/risk-labeled option.
- [ ] It should have stated a safety recommendation: do not store real secrets until OS install, SSH/admin access, TLS, backup, and recovery are verified.
- [ ] It did not explicitly mention Proxmox mutation lock/live-state verification before any Vaultwarden VM changes.

## Context Bloat / Overreach

Observed:
- full active ticket dump generated ~1297 lines and was truncated;
- server change log and multiple runbooks were loaded before a narrow recovery/next-step plan was established;
- no subagent was used for locator/research/review.

## Root Cause Classification

- [x] AGENTS.md/process still not strong enough for context-conserving triage of vague requests.
- [x] Active-ticket overload made priority selection ambiguous.
- [x] Need a script or index for ticket priority summaries.
- [x] Need stronger security-sensitive questionnaire guidance.

## Recommended Fix

- Add/use a ticket summary command/script instead of dumping all active tickets.
- Add a high-priority ticket triage procedure: summarize candidates, select one, then read only that ticket/artifact.
- Prefer subagents for locating/summarizing tickets and related docs.
- For critical secrets/Vaultwarden work, enforce safety gates before real secret migration: OS install, admin SSH, TLS, backups, recovery/break-glass, restore test.
- In questionnaires, label unsafe/risky choices clearly and prefer recommended defaults.