AGENT SKILL

see-a-dashboard

See a dashboard of the side Grafana (tools/observe) rendered before it lands — run the render that starts a throwaway stack from the checkout with fake data, read each dashboard's picture beside the text of its panels, and name a bar with no name, a panel with no data, a query's error or one series where the seed holds several. For a step that changes a dashboard, the collector or Tempo's file, before it signs off, and for one that reviews the change. Writes nothing into the workspace.

License 许可
MIT
所属仓库
emyann/nemia
最近更新
2026-11-15
Stars
★ 3
Forks
0

01 前言 Front Matter

name
see-a-dashboard
description
See a dashboard of the side Grafana (tools/observe) rendered before it lands — run the render that starts a throwaway stack from the checkout with fake data, read each dashboard's picture beside the text of its panels, and name a bar with no name, a panel with no data, a query's error or one series where the seed holds several. For a step that changes a dashboard, the collector or Tempo's file, before it signs off, and for one that reviews the change. Writes nothing into the workspace.

03 SKILL.md 正文

See a dashboard rendered

A dashboard whose queries answer and whose panels draw one nameless bar is not done. The live Grafana shows only what is on main, over what a landed daemon emits: a dashboard, a collector rule or a span that exists only in a worktree is drawn by nothing there. This is how a step sees it anyway: a throwaway copy of the stack, started from the checkout, fed fake telemetry, and read through a picture and the text of every panel.

1. Take the pictures, or read them

  • When a step before you named them in its artifacts, dashboard-<name>.png and dashboard-<name>.txt for each dashboard, read those: the text first, then the picture.

  • Otherwise, and after each change you make, render the checkout you are in:

    node tools/observe/render/render.ts --out <your files directory> [--dashboards steps,time]
    

    It takes a minute and a half and about 0.8 GiB, one at a time on the machine: a second render waits for the first (--wait <seconds>, 600 by default). It needs docker and Playwright's Chromium. It starts a container of its own name (nemia-render-<pid>-<random>) on ports Docker picks, never the live stack, and removes it when it ends, also when it fails. With no --out the files go to $WF_ARTIFACTS, else to .observe/ at the repository's root, which git ignores. Never write them anywhere else in the workspace: a file there rides the branch.

A dashboard is named as its file is, without .json (tools/observe/dashboards/). Name every file the render wrote in your artifacts, by its absolute path: the build's check refuses a change under tools/observe that names no picture and text of each dashboard it changed.

2. Read the text beside the picture

The text is what the page drew, panel by panel, in the order of the page; the picture is the whole page, 1600 pixels wide, in Grafana's dark theme. Read the text first: a panel far down a tall page is small in the picture, and a missing name is easier to see in a line of text.

## Tool errors, across runs [bargauge]
  bar: (no name) = 2
  • bar: <name> = <value> — one line for each bar. (no name) is a bar drawn with no label: the legend names a label the query does not keep, or the query sums the series into one.
  • series: <name> — a chart's legend, one line a series.
  • columns: and row: — a table's header and its rows; only the columns the window shows are written.
  • line: — a log line.
  • no data — the panel drew nothing.
  • error: … — what the panel's query answered, word for word.
  • not seeded: <why> — no fixture meets this panel, on purpose; no data there says nothing about the panel.

Then judge each panel the change touches against what the seed holds:

  • One series where the seed holds several. The seed tells of two steps, build / implement · ada and build / review · agora, in the run seeded-run-1, and every panel that draws series is met by two at least. One bar, or one line, is a query that drops a label or a legend that names the wrong one.
  • A label that is missing. One failed tool_result of the seed carries no tool_name. A panel by tool should show it apart, or leave it out on purpose; a panel that folds it into another bar hides it.
  • no data on a seeded panel. The query names something the seed does not send, or filters it away. If the change adds a span, an attribute or a metric, the seed must send it too: add the fixture in tools/observe/render/seed.ts, whose test (packages/agent-orchestrator/test/observe.test.ts) refuses a name the code does not emit and a panel the seed does not meet.
  • The picture. A panel cut off or overlapping, a unit that reads wrong (4 hours where seconds were meant), a bar chart whose longest label is cut, a legend that says {{step}} as written.

The numbers are fake: read whether they are told apart and labelled, not whether they are plausible. Fixed by hand, the seed can meet a panel and still differ from what the code sends in a value or a unit: the names are held to the code, the values are not.

3. With a browser

A step that has the browser tools can look at the stack itself, to hover a bar, open a legend or change a variable:

node tools/observe/render/render.ts --out <your files directory> --hold [<seconds>]

It renders and writes the same files, then leaves the stack up, 600 seconds unless told, 1800 at most, and prints where Grafana listens and each dashboard's address, on the range, the run and the session the pictures were taken at. Open an address with browser_navigate; there is no login. An edit made in that Grafana is not the checkout's file: load the printed address again to see the file. When you are done, stop it, since it holds the machine's one render slot:

node tools/observe/render/render.ts --stop <the name it printed>

The stack is removed by its own name when its time ends even if you do not. A look in the browser replaces neither the pictures nor the text: both are still owed.

4. Say what you saw

Each thing that is off names the dashboard, the panel by its title, the line of the text or the place in the picture, and what is wrong. A dashboard the change touches with no picture or no text is a finding. If nothing is off, say so, and that you judged from the files: a look that invents work is worse than a short one.

04 同仓库更多 Skills

emyann
emyann/nemia

groom-the-inbox

Read each card in play in the inbox against the repository and say, per card, done (the commit on the base that landed it), replaced (the decision or card that did), stale (each fact that moved, what it said and what it says now) or standing, reading closely only what the period touched. For the janitor's step, which proposes and never acts: a person removes or amends each card by the command the check writes.

★ 3
emyann
emyann/nemia

write-a-spec

Turn an accepted intent into a spec an engineer can build from in this repository: the goal restated, the approach and why, every path it touches, the steps, the tests that prove it, the risks, and what is still open. For a step that designs a change before anyone builds it.

★ 3
emyann
emyann/nemia

review-a-change

Review a change in this repository as a second engineer would: against its intent and spec, by reading the diff and the suite's verdict, with findings that name the file, the line and what goes wrong. For a read-only step that decides whether the work goes round again.

★ 3
emyann
emyann/nemia

read-the-stats

Read a factory's stats since the janitor's last run, against the same length of time before it, write a short report, and propose one correction per card — an allow or deny by name, a line in an agent's body or a skill, or a refusal kept on purpose — each with the numbers behind it, never one already in the inbox. For the janitor's step, which proposes and never acts.

★ 3
emyann
emyann/nemia

land-a-branch

Land a run's result on the base branch in this repository's own checkout, the way the step says it lands: make that one act, read its result, and write the message that records what landed — or change nothing and say why it refused. For a gated ending step (workspace: base).

★ 3
emyann
emyann/nemia

implement-a-change

Build a change from its spec in this repository: tests first where a rule is claimed, small commits, the tests of what changed green for the right reason, and each review finding answered in the next round. For a step that writes the code.

★ 3