agentleFS
maximilianfalcomaximilianfalco/messhall/.claude/skills/messhall-control/SKILL.md

messhall-control

Skill1 starsChanged 34 days agoMITView on GitHub

Verification and control skill for the Messhall repo. Use it to prove a change works before calling it done, to reproduce a vague report, to find how a surface is reached, or before any commit. Triggers on "check as we go", "does it work", "verify", "run it", "reproduce this", "feature map", "what surfaces exist", "before commit", and after building or changing any CLI command, MCP tool, daemon route, feed event or dev command. Pairs a local dev CLI (pnpm messhall-dev) that gathers evidence with a feature map that records every surface, how to reach it, and how to verify it.

The file

We split this skill into the rules it gives an agent

4 rules. About 0 are common, 2 are distinctive, and 2 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.

  • 2

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Run pnpm messhall-dev check.

    line 17
  2. Distinctive:

    Commit. The commit title is the feature, the evidence lives in the reply and the PR.

    line 20
  3. Very specific:

    Verify names a messhall-dev command.

    line 51
  4. Distinctive:

    No pipe characters inside a cell, the parser splits on them.

    line 52

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill 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.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Messhall control

Two parts, used together:

  1. pnpm messhall-dev <command>, a local dev CLI in tools/dev/. Run it instead of writing one-off scripts, so evidence is reproducible across sessions. When it cannot check something you need, extend it rather than working around it.
  2. The feature map, references/feature-map.md. Materialized memory of every surface: how a user or agent reaches it, what it does, where the code is, its status, and the dev command that verifies it. Read it before touching a surface, and update it in the same commit that changes one.

The loop

  1. Find the row in the feature map. If the surface is new, add the row first with status building. Flip a planned row to building when you start it.
  2. Build with tests first (/tdd).
  3. Run pnpm messhall-dev check. It runs format, lint, types, tests and the feature map check in parallel. Nothing ships while it is red.
  4. Exercise the surface with the command in the row's Verify column. Paste the evidence (the table or the response) into your reply and the PR, not a claim.
  5. Flip the row to built, fix its Code and Verify cells, then pnpm messhall-dev featuremap --check.
  6. Commit. The commit title is the feature, the evidence lives in the reply and the PR.

Commands

CommandEvidence it gives
pnpm messhall-dev checkformat, lint, types, tests and the feature map check, as one pass/fail table with the tail of each failed step. Exit 1 on any failure
pnpm messhall-dev featuremap [--check]row counts by status, and every built or building row whose code paths are missing. --check exits 1 on drift
pnpm messhall-dev envnode, pnpm, claude --version and whether it has Channels (2.1.80+), codex --version, the data dir, the daemon's /health on MESSHALL_PORT or 7707, and whether the Claude Code and Codex messhall entries are present
pnpm messhall-dev qa-upload <pr> <files...>commits gifs or screenshots to the qa-assets branch under pr-<n>/ without touching your branch, prints one markdown image per file
pnpm messhall-dev db "<sql>" [--data-dir <d>]a read-only query against a messhall.db, rows as a table. A write fails
pnpm messhall-dev daemon [--keep]messhall daemon from source on MESSHALL_HOME (else a temp dir) and MESSHALL_PORT (else a free port): its url, pid, data dir and /health body. Stops it unless --keep
pnpm messhall-dev room <name> [--data-dir <d>]a room's members with presence and cursor, and its last messages, read only from the daemon's data dir
pnpm messhall-dev store [--data-dir <d>]a scripted join, post, read, done, leave and presence sweep on a scratch room store with a fake clock, then its rooms, members, messages and event counts
pnpm messhall-dev mcp [--tool <name> --input '<json>' --as <role> --room <room>]the MCP server over an in-memory client: instructions, each tool's title, annotations, fields and description length against the budget. With --tool, one call's text and isError
pnpm messhall-dev agent <role> --room <r> [--say <text>] [--wait] [--url <u>] [--key-file <f>]a scripted agent over real HTTP MCP on a running daemon: join, post, wait, read, each reply and how long wait blocked
pnpm messhall-dev feed [--room <r>] [--url <u>] [--since <seq>] [--count <n>]the live SSE feed of the running daemon, read with the human key: a snapshot line, then one line per message, member, presence or room event with its sequence. --since replays from the event log instead. Exits after --count lines or on ctrl-c
pnpm messhall-dev app-shot [--port <n>] [--home <dir>]the Mac app against a seeded scratch daemon on 7796: menu bar label, window, a post made through the app and the daemon-down state, light and dark, as pngs in demo/out/shots/. Look at each before uploading
pnpm messhall-dev demo [--agents 2] [--keep] [--dry-run] [--timeout 240]two real Claude Code sessions in two temp repos agree a contract change through one room: a pass or fail row per step with its time, the room checks, both repo tests, the transcript, tokens per agent and the wall time. Exit 1 on any failed row
pnpm messhall-dev schemawrites contracts/schema.json from the zod contracts (also pnpm schema and pnpm build). A test fails when the committed file is stale

The rest (channel, codex) land with the feature they verify. Their rows in the Dev tool section of the feature map are planned and say what each will prove. Add the dev command before the surface it verifies.

scripts/messhall-dev is a shell wrapper for the same tool, for use from any directory.

Feature map rules

  • One row per surface a user or agent can reach. Columns: Feature, Reach, Does, Code, Status, Verify.
  • Reach is literal: the exact command and flags, the MCP tool name and inputs, the HTTP method and route.
  • Code cells hold backticked repo paths. featuremap --check fails when a built or building row points at a path that does not exist. A planned row names where the code will live.
  • Status is one of planned, building, built, cut. Nothing else parses.
  • Verify names a messhall-dev command. If no command can verify the row, add one to the tool first.
  • No pipe characters inside a cell, the parser splits on them. Write "or" instead.

Gardener rule

When you correct a pattern in review or in your own work, do not stop at the fix. In this order: make the pattern impossible in the codebase (a type, a single paved helper), then a lint rule in oxlint.config.ts, then a line in the plan's "Engineering standards" in personal-dev-notes.md. A comment explaining a workaround is the start of a virus, not a fix.

Download now (ZIP)

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
What did you use it for? e.g. Reviewing pull requests in a Rails app

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.