dogfood-week
Skill46 starsChanged 37 days agoApache-2.0View on GitHub
Score the Agentmetry dogfood week from the live trail and report the beta gate. Use when the user asks about dogfood, the weekly stats, the beta gate, whether a week was green, or how long until beta.
In this repository4 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- outreach-nextSkill
- ship-safeSkill
- trail-debugSkill
The file
We split this skill into the rules it gives an agent
3 rules. About 0 are common, 2 are distinctive, and 1 is 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.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 61
Check the live spool file, not the expired one, before reporting a spool problem.
- Distinctive: line 78
Do not disposition detections on the operator's behalf.
- Very specific: line 80
Only update the private ../agentmetry-enterprise/sales/outreach-log.md if the user asks for the week to be recorded.
≈ “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.
Files
Dogfood week
The beta gate is four consecutive green weeks. It went unstarted for weeks because checking one meant a twenty-minute manual pass, so it never got checked. The point of this skill is to make the question cheap enough to actually ask.
Run
From apps/orchestrator:
.venv/Scripts/python.exe -m agentmetry.cli dogfood
.venv/Scripts/python.exe -m agentmetry.cli stats --days 7dogfood is the authority on the gate. stats gives the seven-day counts and
the per-source breakdown, which is what goes in a weekly note.
Report
State, in this order:
- Consecutive green weeks, out of 4, and the verdict of the most recently completed week
- Event count, sessions, detections, untriaged for that week
- Earliest possible beta close as an absolute date, computed rather than quoted: take the end date of the week currently in progress and add seven days for each green week still needed after it. Never carry a date over from a previous run of this skill. Week 1 of the 2026-08-30 clock went RED, which moved the answer by a week, and every red week moves it again
- Anything RED, and specifically why
- Days left in the week in progress, and its untriaged count. This is the only actionable line in the report. Week 1 closed RED with six untriaged detections after being flagged at three, with three days still on the clock. A report that states the gate without stating the deadline is why
A week is green when the recorder ran on at least three days, the trail chain verifies, every critical or high detection was dispositioned, and nothing is stuck in the hook spool.
What counts as a problem
- Untriaged critical or high detections. This is the usual cause of a red week and it is fixable by the operator in minutes. Say which detections.
- Fewer than three active days. This is the one that genuinely cannot be fixed retroactively, because no disposition creates a day the recorder did not run. Untriaged detections are the opposite: always fixable, at any time.
- A moved ruleset fingerprint. This restarts the clock. Check it against
the value in
CLAUDE.md, which is the one place that number is written down, and say so loudly if it has moved. This file used to carry its own copy. It went stale, and a skill that hardcodes a hash it does not own would have reported a restarted clock that had not restarted.
hook-spool.expired.jsonl is not a stuck spool. Expired entries can never
drain by definition, and the dashboard warning does not distinguish the two.
Check the live spool file, not the expired one, before reporting a spool problem.
Rules
- Never edit detection rules, traits, the engine, or the MITRE mapping. The
freeze has no fixed end date. It runs until four consecutive green weeks have
closed, so a red week extends it. Editing any of those files restarts the
clock this skill exists to measure. See
CLAUDE.md. A red week from untriaged detections CAN be rescued, at any time. Every verdict is recomputed on every run:
build_reportreads the disposition store live (core/audit/dogfood.py:289), so nothing is frozen when a week ends. Week 1 of the 2026-08-30 clock scored RED for four days and flipped to GREEN the moment its six detections were dispositioned on 2026-09-09.This file previously said the opposite, in bold, and was wrong. The claim came from reading the RED verdict and assuming it was final instead of reading the code. If an operator is looking at a red week, tell them to triage it.
- Do not disposition detections on the operator's behalf. Triage is a judgement about their own machine. Report what needs triage and let them decide.
- Only update the private
../agentmetry-enterprise/sales/outreach-log.mdif the user asks for the week to be recorded.
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.

