Skip to content
Draft
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 1 addition & 1 deletion .agents/skills/quest-afk/SKILL.md
Original file line number Diff line number Diff line change
@@ -1,6 +1,6 @@
---
name: quest-afk
description: Keep this session's work going while the user is away, taking your recommendations and deferring merge picks until they return.
description: Keep this session's work going while the user is away, taking your recommendations and deferring merge picks until they return. Use when the user says afk, afk merge, or afk soon.
---

Run `quest skill afk` and follow its output.
Expand Down
2 changes: 1 addition & 1 deletion .claude/skills/quest-afk/SKILL.md
Original file line number Diff line number Diff line change
@@ -1,6 +1,6 @@
---
name: quest-afk
description: Keep this session's work going while the user is away, taking your recommendations and deferring merge picks until they return.
description: Keep this session's work going while the user is away, taking your recommendations and deferring merge picks until they return. Use when the user says afk, afk merge, or afk soon.
allowed-tools: Bash(quest skill afk) PowerShell(quest skill afk)
---

Expand Down
15 changes: 10 additions & 5 deletions README.md
Original file line number Diff line number Diff line change
Expand Up @@ -244,13 +244,16 @@ second name in one `## Audit`, and `## Skipped audits` outside the root.

A quest waiting outside the tree says so in a `## Waiting` section, one item
per condition, each opening with who acts: `Maintainer:` for a decision or
action only the maintainer can take, or the outside party, such as
`moq-dev/moq:`. `quest check` rejects an item that names nobody.
action only the maintainer can take, `Maintainer (PC):` for one they can take
only at one of their computers, such as a local login or a hardware check, or
the outside party, such as `moq-dev/moq:`. `quest check` rejects an item that
names nobody, and a maintainer qualifier other than `(PC)`.

```markdown
## Waiting

- Maintainer: open a Hetzner account; `GET /server` on the Robot API answers 200
- Maintainer (PC): plug in the capture card; `v4l2-ctl --list-devices` lists it
- moq-dev/moq: a release carries the VA-API bindings
```

Expand All @@ -261,8 +264,10 @@ body.
milestone and the same downstream impact `quest unblocked` counts; a draft
ranks as the quest its branch carries. Each line gives the dependents, whether
the quest is otherwise ready or still blocked, its path or the draft's URL, and
the item; `--maintainer` keeps only the maintainer's, and `--json` prints
`{path, pr, dependents, ready, waits}` objects. A draft's item that names no
the item; `--maintainer` keeps only the maintainer's, `--no-pc` drops the
`Maintainer (PC):` ones for a session away from your computers, and `--json`
prints `{path, pr, dependents, ready, waits}` objects, each wait's `pc` marking
those. A draft's item that names no
party is the maintainer's. Drafts come from
the upstream's open PRs, found as for `unblocked --unclaimed`; without the
remote, off GitHub, or when GitHub cannot be read, it lists the tree's items
Expand Down Expand Up @@ -2465,7 +2470,7 @@ arrives with the guide sections and other skills it builds on, in one call:
| `quest-review` | Review a PR's branch locally with a model other than its latest writer, then mark it ready for review. |
| `quest-merge` | Queue a PR to land, running `quest-iterate` first when it needs work. |
| `quest-run` | Orchestrate a project from one session: start ready quests, take PRs to merge, and settle what quests and drafts wait on you, in a loop until only your decisions remain, then plan the follow-ups no quest owns yet. Pass a filter such as `prs` to start no quests. |
| `quest-afk` | Keep a session such as `quest-run` going while you are away: it acts on its recommendations, defers merges, and walks you through the rest when you return. |
| `quest-afk` | Keep a session such as `quest-run` going while you are away: it acts on its recommendations, defers merges, and walks you through the rest when you return. `afk merge` also merges low- and medium-risk PRs it strongly recommends, and `afk soon` asks pending questions in one batch before you leave. |

Use `/quest-plan` in Claude Code or `$quest-plan` in Codex. Skills coordinate
within your agent session. A started quest stops at a pull request; merging is
Expand Down
1 change: 1 addition & 0 deletions assets/AGENTS.md
Original file line number Diff line number Diff line change
Expand Up @@ -83,6 +83,7 @@ An epic carries no size until its last child is deleted; its README is then the
A quest without it runs anywhere; `quest unblocked --json` reports the list as `hosts` so a dispatcher routes work without reading prose.

`Waiting` lists what the work waits on outside the tree; each item opens with who acts, `Maintainer:` or the outside party, before a colon.
Open a step the maintainer can take only at one of their computers, such as a password-manager unlock, a hardware check, or a local login, with `Maintainer (PC):` instead, so a session away from them skips it; a decision or a step on a website stays `Maintainer:`.
A decision item gives its options and a recommendation.
Remove an item once its condition clears, and the heading with its last item.

Expand Down
17 changes: 12 additions & 5 deletions assets/skills/afk.md
Original file line number Diff line number Diff line change
@@ -1,28 +1,35 @@
---
description: Keep this session's work going while the user is away, taking your recommendations and deferring merge picks until they return.
description: Keep this session's work going while the user is away, taking your recommendations and deferring merge picks until they return. Use when the user says afk, afk merge, or afk soon.
guide: Summaries
---

The user has stepped away and will not answer.
The user has stepped away and will not answer; a message of just `afk`, or `afk` and a mode below, runs this skill.
Keep this session's work going within its scope: the skill it is running, such as `/quest-run`, or else the task at hand. In a session with no work under way, run `/quest-run`.
Ask nothing: wherever you or an agent would ask the user, take your recommendation instead, unless the decision is deferred below.
A deferred question counts as asked, so a skill's loop goes on past it and finishes as that skill describes.
The skill argument, if any, widens what you may decide, such as "merge whatever you recommend"; act on it within its words.

The argument picks a mode, and modes combine, such as `soon merge`:

- None: every merge pick below waits for the user.
- `merge`: also run `/quest-merge` on each PR you strongly recommend merging that `quest status` shows low or medium for its head, with a review that counts for that head, no open decision, no follow-ups, and no quest deleted or deprioritized beyond the one it finishes. Defer every other PR, a high or unclassified one included.
- `soon`: the user is about to leave. Ask every question the session has pending now, in one batch, as `/quest-run`'s "decisions" does, and act on the answers; once the prompt returns, go on as afk, taking or deferring each question it left unanswered as below.

Other words widen what you may decide, such as "close superseded PRs"; act on them within their words. A grant of merges that names no risk tier is `merge`.

Take every decision a later PR review can still undo:

- On a quest or PR this session works on itself, take your recommendation on each decision, list those choices in the PR body for review, and run `/quest-review` when nothing else remains.
- Tell each agent you dispatch that the user is away, so it does the same.
- As a coordinator, dispatch `/quest-iterate` on a PR with open decisions, with your recommended answers in its brief.

Unless the argument allows it, defer as blockers what no review can undo and what only the user can do:
Unless the mode or argument allows it, defer as blockers what no review can undo and what only the user can do:

- Merge picks beyond a standing `/quest-merge` pick, deleting or deprioritizing a quest, closing a PR, and taking over a stale claim or hold.
- A follow-up's decisions, which wait for its PR's pick.
- `/quest-plan` sessions, and actions only the user can take, such as granting access or providing a secret.

Carry on each pick the user already made until it lands or needs them again.
When the work finishes, end with a final summary holding only the blockers, each linked, with its options and your recommendation, so a fresh session can act on it; the choices you took live in the PRs' bodies.
When the work finishes, end with a final summary holding only the blockers, each linked, with its options and your recommendation, so a fresh session can act on it; list the `Maintainer (PC):` steps, which need the user at a computer, apart from the rest. The choices you took live in the PRs' bodies.

A message that adds to these instructions, such as a newly granted merge or more work, amends them, and afk continues.
Any other message is the user's return: afk ends. Answer it, then resume the session's skill, or `/quest-run` on the blockers when it ran none, which asks the questions it deferred.
7 changes: 4 additions & 3 deletions assets/skills/run.md
Original file line number Diff line number Diff line change
Expand Up @@ -67,6 +67,7 @@ Rewrite a quest that waits in prose into a `## Waiting` section.
Check each condition yourself, in parallel sub-agents.
An outside party's item that holds stays, asked about only when the user can advance it, such as by requesting an upstream release.
Ask each maintainer item left among the round's questions, in that order, saying what it unblocks; a skip stands for the session.
A `Maintainer (PC):` item needs the user at one of their computers: when the argument says they are away from one, such as `remote` or `on my phone`, read the waits with `--no-pc` and leave those items for a later session; when it says they are at one, such as `at my PC`, ask those first.
For an action only the user can take, give the exact steps and the check you will run once they report it done; take any secret through the host's secret prompt.

Settle a draft's items with its PR, naming cleared ones in its `/quest-iterate` brief.
Expand All @@ -76,11 +77,11 @@ Make these edits in one draft PR on a `quest-run/<slug>` branch, run `quest chec
## Loop

Each round, rerun `quest unblocked --unclaimed --json` and refresh the open PRs, since merges and released claims make more work ready.
When no agent is running and every question has been asked, block with `quest wait` on the queued PRs whose merge would make a quest ready, and run a round once one lands or leaves the queue.
Never wait on any other PR, and read none between rounds: list each PR left for a later round in the summary, with its pick.
When no agent is running and every question has been asked, block with `quest wait` on the queued PRs whose merge would make a quest ready, and on each PR ready for review whose checks are still running and which a round would then pick or ask about; run a round once one lands or leaves the queue, or its checks pass or fail.
Wait on no other PR, such as a draft or a skipped one, and read none between rounds: list each PR left for a later round in the summary, with its pick.
A usage or rate limit is waited out until its reset; it never ends the run.

Finish once nothing is running, no queued merge would make a quest ready, every PR left needs the user or a later round, and every wait in scope is decided, skipped, or left to an outside party.
Finish once nothing is running, no queued merge would make a quest ready, no PR waited on above still has checks running, every PR left needs the user or a later round, and every wait in scope is decided, skipped, or left to an outside party.
Then run `/quest-plan` on the run's follow-ups, each in a checkout of the repository whose quest tree owns it: problems it surfaced that no quest or open PR owns, such as a blocker agents hit, and the follow-ups kept in each of this run's own merge picks for another repository's quest tree, landed or not, unless the user withdrew the pick.
Write the final summary, naming those plan PRs and listing the waits settled, cleared, or deleted and each still held with who acts and its check's result, so a fresh session can act on it, with each open decision's options beside your recommendation.

Expand Down
41 changes: 0 additions & 41 deletions quest/m0/afk-and-waits.md

This file was deleted.

2 changes: 1 addition & 1 deletion quest/m1/cursor-reviewer.md
Original file line number Diff line number Diff line change
Expand Up @@ -61,5 +61,5 @@ Guidance:

## Waiting

- Maintainer: log Cursor's CLI in on a macOS host of the maintainer's, at a version no older than `2026.11.13-eacade1`, so an agent can run the isolation test on macOS before this quest's PR merges. Check with `cursor-agent status`, which prints `Logged in as <email>` when done. An agent cannot finish Cursor's browser login on the maintainer's account; it can drive the maintainer's agent browser through `NO_OPEN_BROWSER=1 cursor-agent login`'s URL once the maintainer picks the account, as it did on Linux. The maintainer's Linux workstation that runs agent sessions was logged in on 2026-11-21 (CLI `2026.11.19-f005e9d`, from nixpkgs' unfree `cursor-cli` through `nix shell`, not installed); until kixel-ai/dotfiles installs it on the hosts, the implementer gets it with `NIXPKGS_ALLOW_UNFREE=1 nix shell --impure nixpkgs#cursor-cli`.
- Maintainer (PC): log Cursor's CLI in on a macOS host of the maintainer's, at a version no older than `2026.11.13-eacade1`, so an agent can run the isolation test on macOS before this quest's PR merges. Check with `cursor-agent status`, which prints `Logged in as <email>` when done. An agent cannot finish Cursor's browser login on the maintainer's account; it can drive the maintainer's agent browser through `NO_OPEN_BROWSER=1 cursor-agent login`'s URL once the maintainer picks the account, as it did on Linux. The maintainer's Linux workstation that runs agent sessions was logged in on 2026-11-21 (CLI `2026.11.19-f005e9d`, from nixpkgs' unfree `cursor-cli` through `nix shell`, not installed); until kixel-ai/dotfiles installs it on the hosts, the implementer gets it with `NIXPKGS_ALLOW_UNFREE=1 nix shell --impure nixpkgs#cursor-cli`.
- kixel-ai/token-max: `token-max sessions` prices Cursor CLI sessions, so `quest eval run` reports Cursor reviewers' real cost instead of a lower-bound `$0.00`. Check that a Cursor review session's line in `token-max sessions` shows a nonzero cost. Advance it with `/quest-plan` in that repository; its tree had no such quest on 2026-11-17. If that stalls, the Plan's fallback prices Cursor seats in `quest eval` from the usage its stream reports.
Loading