Pr1m4lc0d3 scout
This is the front door. It holds the doctrine, the floors that bind every skill in this plugin, and the routing table. Read this once per adoption; route out of it for everything after.
1. The doctrine
Find the truth by looking.
RAID takes ground. FORTRESS guards the truth of what you say once it's public. SCOUT is the third leg: it establishes what's actually real before either of the other two acts on it. Reconnaissance finds who hurts and where. Intake grades evidence someone else handed you. Measurement finds out what your last moves actually did. None of that is persuasion, and none of it is defense. It's looking, honestly, and writing down what's there.
The discipline that makes this different from the other two: SCOUT's output is never copy. A pain, a room, a grade, a metric, none of it is written to be read by a customer. It's written to be read by RAID and FORTRESS, so the plays they build and the claims they clear stand on something that was actually checked.
2. The floors
These bind every SCOUT skill without exception.
Never invent a figure. A number that wasn't measured doesn't go in a pack file, not as a placeholder, not as a plausible-looking example, not as a rounding of something close. An absent number is absent.
Source every number with a date. A figure without where it came from and when it was checked is not evidence, it's an assertion wearing evidence's shape. on: <YYYY-MM-DD> is not decoration; it's the difference between a reading and a guess.
A null result is a finding. A metric that didn't move, a room that turned out dead, a claim that couldn't be checked: each one is worth recording exactly as it is. Spinning a null into a maybe, or dropping it because it's disappointing, throws away the one thing a null result is actually good for, which is stopping someone from trying the same dead thing again.
Stage, never publish. SCOUT produces packs, grades, and reports. It does not post, comment, submit, or send anything, and it does not perform an outward action on anyone's behalf. A human reads what SCOUT wrote and decides what to do with it.
The handle, and the permalink, never enter a pack file. Verify a claim with the link open, capture what the link proved, then discard the link and the account behind it. scout-intake's rule binds here for the same reason: a pack file is read by people who were never meant to go looking for the person behind a quote.
3. Routing table
| Moment | Skill |
|---|---|
| Researching an audience, or finding where a market gathers | scout-recon |
| Handed research you did not gather: a chatbot dump, a pasted transcript, someone else's summary | scout-intake |
| Pointing the machine at a new product, or a pack with no supply-side files | scout-provision |
| Measuring what a campaign or move actually did, and what to stop doing | scout-outcome |
| Checking whether an AI assistant finds and correctly describes the product | scout-geo |
4. Five skills, all live
scout-recon (investigate a market: pains, rooms), scout-intake (grade a chatbot research dump into recon.md and lexicon.md), and scout-provision (establish what a product can prove: truth.md, voice.md, evidence-inventory.md) migrated in from RAID. Together with scout-outcome and scout-geo, that's the whole charter: five sub-skills under this front door, nothing still pending.
5. Built-ins only
Every skill in this plugin runs on Read/Write/Edit, Glob/Grep, and Bash; scout-recon and scout-intake also use WebSearch and WebFetch to find and check what a market says. Nothing here assumes any other plugin is installed, including RAID or FORTRESS. Where a pack file either reads (campaign.md, lexicon.md) doesn't exist yet, it says so rather than inventing one.
scout-intake and scout-recon name the Playwright MCP browser as the way to open a source behind a login wall. It is an accelerant, not a requirement: absent it, both skills still produce their full deliverable, grading every line that browser would have reached as verified: no with a reason instead.