panal
AGENTS.md1 starsChanged 45 days agoApache-2.0View on GitHub
Claude Code, Codex, opencode and any other agent working here reads this file. Read it before making changes. Panal (panal, Go, Bubble Tea) is a read-only terminal dashboard of what opencode, codex, agy and Claude Code as orchestrator are doing, plus panal delegate (hands tasks to those CLIs with quota fallback) and a router (-c auto) that learns which agent, model and effort fits each task. Overview: README.md; guides: docs/; structure: docs/ARCHITECTURE.md. 1. Work only inside this repository (or its…
- Installs packages
- Commits and pushes
Where files are headingNew AGENTS.md files have more rules about keeping secrets out of the code, and fewer about code style.
Of the AGENTS.md files we read, the 33,315 first seen in the last six months, against the older ones: the kind of rule that gained most share, and the kind that lost most. About files like this in general, not this one, and an estimate, since a kind is a pattern in the words.
In this repository4 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- llms.txtllms.txt
- panal-uxSkill
- tui-designSkill
The file
We split this AGENTS.md into the rules it gives an agent
5 rules. About 0 are common, 2 are distinctive, and 3 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈2
Distinctive · estimate
This author's own choices. Worth a look.
3
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 18
Don't edit or commit to any other project of the user, including the repo where delegar.sh lives.
- Very specific: line 86
Golden screens are reviewed: After -update, read the diff: only the intended change may appear.
- Distinctive: line 110
No personal data; no AI attribution in commits.
- Very specific: line 115
Before touching internal/ui or internal/mascots, read .claude/skills/panal-ux/SKILL.md: palette, status glyphs, mascot and animation rules, …
- Distinctive: line 124
Read it when you pick the project back up, and add an entry when you finish a work session.
≈ “Common” is an estimate: at least 1 in 5 of the 42,283 AGENTS.md files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Rules for agents in this project
Claude Code, Codex, opencode and any other agent working here reads this file. Read it before making changes.
What it is
Panal (panal, Go, Bubble Tea) is a read-only terminal dashboard of what
opencode, codex, agy and Claude Code as orchestrator are doing, plus
panal delegate (hands tasks to those CLIs with quota fallback) and a
router (-c auto) that learns which agent, model and effort fits each
task. Overview: README.md; guides: docs/; structure:
docs/ARCHITECTURE.md.
Rules
- Work only inside this repository (or its git worktrees).
Don't edit or commit to any other project of the user, including the repo
where
delegar.shlives. If an improvement would need a change outside this repo, solve it on the dashboard's side, or tell the user and stop. - The repo is public at https://github.com/AlbertoVasquezR/panal (module
github.com/AlbertoVasquezR/panal). Local commits onmaster(or on the worktree's branch) are fine;git push, pull requests, tags and releases only when the user asks for them. - Because it is public, never put personal paths, email addresses, tokens or names of work projects in code, tests or docs.
- No
Co-Authored-Byor any other AI attribution in commit messages. - English everywhere: code, identifiers, comments, UI text, commit
messages and docs. The product name is Panal in prose and on screen;
the command, binary, module, flags, file names and config keys stay
panal.
Commands
go build ./...
go vet ./...
go test ./...
gofmt -l . # must print nothing
go test ./internal/ui -run TestScreens -update # regenerate golden screens, then read the diff
go run ./cmd/panal -preview 100 # one screen at 100 columns (add -screen history, …)
go run ./cmd/panal -doctor # sources found, what was read, models
go install ./cmd/panal # installs the binary to ~/go/binRepo map
| path | owns |
cmd/panal | flags and subcommand dispatch |
cmd/preview | every mascot frame to a PNG |
internal/state | the shared Row, Status, Quota |
internal/readers | one reader per agent; read-only; real samples in testdata/ |
internal/live | live quota queries to the CLIs (no turn, no quota spent) |
internal/runs | run file format and ~/.panal paths; legacy.go reads the delegar.sh format |
internal/history | past runs with tokens, cost, ratings, full tasks |
internal/delegate | the delegate, route and models commands; each CLI's command line; quota patterns |
internal/router | tier rules, outcomes, posteriors, Thompson choice, external router |
internal/models | model discovery, models.json, cost heuristic, pool = auto |
internal/feedback | panal feedback and feedback.json |
internal/changes | what a run really touched (git, read-only) |
internal/activity | last action, loops and tests from a log tail |
internal/credits, internal/forecast | codex credits; when a quota runs out |
internal/report | task types, -report, -summary |
internal/remote | -serve and other machines |
internal/config | panal.conf; which agents are off |
internal/alerts | notifications, bell, ntfy, webhooks |
internal/mascots, internal/ui | sprites; Bubble Tea views and keys |
internal/pet | panal pet: the mascot for a split pane and its JSON frame contract (docs/reference.md#pet-frames) |
plugins/panal-pet, .claude-plugin/marketplace.json | the Claude Code plugin that draws panal pet -stream frames above the prompt (TypeScript hooks module; check with claude plugin validate) |
Invariants
- The dashboard is read-only. Panal never writes to the agents' files.
It writes only its own files under
~/.panal(PANAL_DATA): run files and logs,samples.json,feedback.json,router-checks.json,models.json. - External formats keep their names. JSON keys, paths and values that belong to codex, agy, opencode or Claude Code are never renamed.
- The legacy delegar.sh format (Spanish keys) is read only in
internal/runs/legacy.go. - Spanish only where allowed:
internal/runs/legacy.go, regex patterns that match tasks written in Spanish (each one says so), and real captured output intestdata/. - No real agents in tests. Fakes only (the test binary as a fake CLI, real captured outputs); never spend quota.
- Golden screens are reviewed. After
-update, read the diff: only the intended change may appear. - The router stays explainable. Tier rules live in one table with names
that
panal routeprints; Thompson sampling is seeded in tests.
Where to change X
| to… | change | and also |
| read a new agent | a reader in internal/readers (implements Reader), added to All() and Sources() | real samples in testdata/; a color and a mascot (read the UX skill first) |
run a new CLI from panal delegate | agents in internal/delegate/agents.go, its patterns in classify.go, DefaultCLIs in chain.go | a fake in fake_test.go; config.Known; discovery in internal/models/discover.go |
| add a key | the view's key switch in internal/ui/ui.go, the help in help.go, the footer | docs/dashboard.md#keys; golden help screens |
| add a router tier rule | tierRules in internal/router/tier.go | tests in English and Spanish in tier_test.go; the table in docs/router.md |
| change outcomes or priors | internal/router/evidence.go, posterior.go | seeded learning tests; the numbers in docs/router.md |
| add a config key | Conf and Read in internal/config/config.go | docs/configuration.md |
| add a run file field | runs.Run in internal/runs/runs.go (with omitempty) | docs/reference.md#run-files |
Definition of done
gofmt -l .prints nothing;go build ./...,go vet ./...andgo test ./...pass.- New behavior has tests; UI changes have reviewed golden diffs.
- The docs in
docs/(and the README, if it is user-facing) say what the code does; every command, flag, key and path they mention exists. - No personal data; no AI attribution in commits.
- An entry in
docs/JOURNAL.mdwhen you finish a work session.
UI design
Before touching internal/ui or internal/mascots, read
.claude/skills/panal-ux/SKILL.md: palette, status glyphs, mascot and
animation rules, sizes and checklist. For general TUI principles there is
.claude/skills/tui-design/ (a copy of gfargo/tui-design-skill). What is left
from the last UX audit is in docs/ux-audit.md.
History
What was done, why, what was learned and what is pending: docs/JOURNAL.md.
Read it when you pick the project back up, and add an entry when you finish a
work session.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

