lux/docs/constellation-work.md

# Constellation: working state

Read [the design](constellation.md) and this file to resume. Keep this file short:
replace superseded state rather than appending a chronological log.

## Working agreement

- Explore the philosophy through concrete use, discuss meaningful choices, then
  implement a small coherent change. Curiosity and alternative designs are welcome;
  an idea need not prove a defect before we explore it.
- Write succinctly without losing important details. Distinguish accepted choices,
  proposals and observations. Ask when intent would change the design; resolve
  routine implementation details autonomously. Infer opportunities from the user's
  philosophy and actual use; contribute design ideas beyond literal backlog items.
  Continue to the next natural step when the user could only usefully say "continue";
  pause for material choices or missing authority, not routine progress.
- Minimize input stops: plan a coherent batch of autonomous work and gather the
  material inputs needed for the next batch together. Avoid serial question/action
  loops. When pausing, state exactly what input is needed and what it will decide;
  otherwise continue with the next logical step rather than ending with an offer.
- The main agent owns alignment, design and integration and may do small work.
  Use subagents for bounded independent investigations or implementation with clear
  ownership. Begin with at most two workers; avoid overlapping writes, duplicate
  investigations and mandatory delegation of trivial tasks.
- Give workers the question, relevant paths, constraints and expected result.
  Prefer a compact brief and fresh context when sufficient; retain relevant prior
  context for follow-ups. Return conclusions, uncertainty and evidence pointers,
  not raw logs. Astra/Sol orchestration is a candidate to learn from, not a fixed rule.
- Read only what can affect the current decision. Use quiet commands and bounded
  output; keep large logs in `/tmp` and inspect relevant failures. Do not repeatedly
  check settled facts. Consult current documentation for consequential or unfamiliar
  interfaces when needed for the task.
- No tests unless explicitly requested. Keep other verification proportional;
  report only what was actually checked. Avoid turning small changes into campaigns.
- Keep durable decisions in the design and current questions/next steps here.
  Add an area-specific document only when an active topic outgrows this arrangement.
  No separate backlog, scratchpad, audit ledger or agent-role files for now.

## Boundaries

Repository remotes are read-only. Only the user activates system generations.
Agents must not have write access to live company systems; test writes require
task-specific authority and care. Do not read secret values for layout discovery.
Current exploration does not authorize relocating credentials, synchronizing
company content or deleting existing material. Existing local changes are preserved.

## Git baseline — 2026-09-09

The accumulated NixOS changes are recorded in focused local commits on
`chore/workflow-system`: AI tooling/inputs, editor ergonomics, query persistence
and target context, workspace navigation, Notes, SharePoint, Zellij and TeamViewer.
The original long handoff and archived plans are preserved in commit `4e3a3ac`;
use Git history for chronology. `nvim.log` is retained locally and ignored.
Workbench changes are committed on its local `main`: `eb8da6a` organizes HTTP/SQL
material and `063157c` archives helpers and documents native tools. The ordinary
company-root files are outside either repository and remain user-owned.
This cleanup preserves implementation behavior; a commit does not establish runtime
acceptance. Keep subsequent implementation and its documentation in the same
coherent commit before starting unrelated work.

## Implemented state

- **Places:** `~/yookr` has ordinary `.envrc` and `AGENTS.md` files; its environment
  loads workbench tools only. Workbench is at `~/yookr/workbench`, company projects
  at `~/yookr/projects`, and personal groups at `~/projects/{own,ext,forks}`.
  Direct moves retained archives/worktrees; all three linked dashboard worktrees
  were repaired. Nix owns tools/integration, not these workspace files.
- **Navigation:** Walker enters Yookr or a repository; Git identity follows the
  company root; XDG projects points to `~/projects`. Notes has a desktop entry
  opening `~/sync/notes` without creating a daily note. Existing note commands,
  project search and Niri daily-note behavior remain.
- **Synchronization:** Bruno collections were previously moved to Trash on Zenith
  and their shared Syncthing entry removed. Other devices' files were untouched.
  Syncthing was subsequently observed running. Notes and credentials remain in place.
- **API/database work:** workbench `http/` groups requests by service with shared
  authentication; `databases/` groups reusable queries by database. Private HTTP
  environment files are ignored at any depth. Dadbod working queries persist under
  `stdpath('data')/db_ui/working`; write them there without promoting them into Git.
  Saved-query storage also contains connection state and must stay outside Git.
  Lualine uses native Kulala environment and Dadbod connection/timing components.
  These indicators are not access controls.
- **Editor ergonomics:** word-boundary wrapping, count-aware displayed-line `j/k`
  in ordinary normal/visual buffers, and Blink Escape cancellation. The existing
  `tests/neovim-ergonomics.lua` is retained; earlier smoke-check evidence is in the
  archived plan, with interactive preview restoration/visual behavior unproven.
- **Agents:** agent-deck, Herdr and tmux coexist; no replacement has been chosen.
  Codex has agent-deck notifications and hooks enabled. Shared Context7/NixOS skills
  live under `.agents/skills`; a Nix-managed marketplace exposes Ponytail.
  Zellij is also declared on workstations. None of this selects a session workflow.
- **SharePoint:** Zenith declares an hourly service syncing `company-sharepoint:`
  into `~/.local/share/yookr-reference/sharepoint/Algemeen` and
  `company-sharepoint-projecten:` into `Projecten`. Each invocation has
  `--max-delete 100`; new files use `UMask=0077`. The latest implementation only
  checks for `rclone.conf`; the old environment-file gate is gone. Archived setup
  instructions describe an earlier version. Consent, remote configuration and
  synchronization state were not inspected during this cleanup.
- **Remote support:** Zenith enables TeamViewer. Temporary Xorg/Xfce settings are
  comments only. The purpose beyond remote support and runtime acceptance were
  not recorded; preserve the change without calling it a settled design.
- **Inputs:** accumulated lock updates are retained, including an older Elephant
  revision (`376ee71c` replacing `23f37238`). No rollback rationale was recorded.

## Activation and verification

Earlier handoffs report user activation of the relocated XDG/Git/Walker paths and
Dadbod working-query persistence. Notes and the newer target statusline were last
recorded as pending activation. Only Zenith's filesystem was relocated; other hosts
need their own migration before adopting the shared paths. Reopen shells or
recreate path-dependent development environments as needed; no compatibility links
or cache rewrites were introduced.

For this cleanup, `nix flake check --no-build --quiet` and diff whitespace checking
passed in both NixOS and workbench. NixOS emitted the expected dirty-tree and unknown
`modules` warnings; workbench emitted only the dirty-tree warning.
No builds, tests, activation, credential reads or company-service calls were
performed. Active-generation differences were not checked; earlier activation
reports are not a fresh comparison with this baseline.

## Open choices and next discussion

The focus is the broader environment, not further API/database polish. Candidate
areas remain places/ownership, navigation/interaction, human/agent collaboration,
and system operation/maintenance; these are discussion areas, not a work queue.

Current discussion: consistent project/agent work and Niri workspace use. The user
values Herdr's spaces and overview of running/waiting agents, but dislikes its
keybindings and mouse orientation. Agent-deck has startup issues (not diagnosed).
Kitty/Niri, tmux and Zellij are candidates, not selected replacements.

Read-only inspection found existing Niri overview, workspace movement and tabbed
columns; Kitty has no declared project/session organization. Installed versions:
Kitty 0.48.2, Niri 26.04, tmux 3.7c, Zellij 0.45.1 and Herdr 0.9.0.
Herdr's installed default config confirms remappable navigation and agent jumps.
Current [Herdr keyboard docs](https://herdr.dev/docs/keyboard/) describe direct
chords; [Kitty sessions](https://sw.kovidgoyal.net/kitty/sessions/) supply native
session creation/switching. Session layouts and window overviews do not themselves
provide authoritative agent attention states. tmux detach keeps processes alive;
Zellij resurrection recreates layouts/commands, not process memory. None of these
capabilities was exercised interactively; no sessions were inspected or changed.

Proposed workflow, awaiting discussion: organize human-visible work by project/task;
one lead conversation owns each coherent change. Use bounded subagents when work
is independent, with explicit ownership and evidence returned to the lead. Separate
worktrees for independent writing efforts, one integration owner. Keep enduring
repository rules in AGENTS.md, current goals/acceptance in task prompts, and durable
decisions/current state in existing project docs. No fixed agent team is selected.

Proposed comparison: hold the desktop and model settings fixed; compare solo,
solo plus independent review, then bounded parallel delegation on the same starting
commit and task. Record accepted correctness, unnecessary complexity, human
interventions/rework, total completion time and aggregate usage including workers.
Use fresh isolated runs and common acceptance checks; repeat promising variants
before drawing reliability conclusions. Measure desktop navigation separately.
[Codex subagents](https://learn.chatgpt.com/docs/agent-configuration/subagents)
and [evaluation guidance](https://developers.openai.com/api/docs/guides/evaluation-best-practices)
support bounded delegation and task-specific evaluation; the proposed trial is
not a measured result. No benchmark framework or comparative runs created.

Inputs requested together: typical project/agent count and clients; required
persistence (project switch, terminal close, logout/reboot, remote connection);
a representative trial task and rough time/spend limit. These determine the first
concrete trial. No configuration changes, tests, activation or agent runs in this
discussion; documentation whitespace checked.

Mail search remains unresolved. The workbench's custom mail reader, inventory
pilot, justfile and old README are archived; native Garden/kubectl/K9s now own the
documented operations. No replacement mail helper or semantic index is selected;
raw mailbox grep is not assumed sufficient. No new ingestion is authorized.

Human credential delivery remains both test/live via environment variables with
ordinary editor startup. Workbench `.envrc` supplies aggregate credentials; agents
must not source it. Separate read-only agent credentials and access isolation are
explicitly deferred, not prerequisites for unrelated work. Do not resume the
rejected per-connection launchers or per-selection decryption proposal.

Notes placement and further company material organization remain open. Work is
mostly project-based and notes mostly meetings; do not impose investigation
folders, outer project containers or company-wide editor search without a need.