intel-discipline-advisor skill

Intel Discipline Advisor

Purpose

Triage a competitive or market question into the right intelligence response: which of the eight collection disciplines to run, on what cadence, feeding which artifact, executed by which skill. The intelligence-collection-disciplines compendium holds everything about every channel; this advisor answers the question a busy PM actually has — "given what's on my desk, which two or three channels matter, and where do my limited hours go?" Running every discipline on every question is the failure mode; scoping to the decision is the craft. The advisor teaches the mapping as it routes, so by your third session you won't need it. That's the goal.

Input

Works best with: the decision or question on your desk, in your words — "I think [Competitor A] is building something," "my TAM slide got shredded," "sales keeps getting surprised." Also useful: any signal you've already noticed (a job posting, a pricing change, an earnings remark), your time budget, and whether you're limited to free sources.

Anything supplied with the invocation itself — text after the skill name, a pasted context dump, or an appended ARGUMENTS: line — counts as answers already given. Use it and skip whatever it covers; don't re-ask.

Arriving empty-handed? That works too. The advisor opens by asking what's on your desk, with enumerated situations to pick from.

Example invocation: Intel discipline advisor: two of their senior engineers just followed our CTO on a preprint server, and their careers page doubled — what do I run?

Key Concepts

default interaction protocol (entry modes, one question per turn, progress labels, numbered recommendations). This file defines the domain logic.

• Facilitation protocol: use workshop-facilitation as the

every PM artifact has a known discipline mix and refresh cadence. The advisor's job is matching the user's situation to a row — and showing the match, because the mapping is the lesson.

• The routing brain is the artifact-mapping table in intelligence-collection-disciplines:

already flagged once — the recommendation starts from "1 discipline flagged" on the confidence stacking ladder and names which independent channels would corroborate (see autonomous-investigation).

• Signals already seen are a head start. If the user noticed a job posting surge, HUMINT has

releases change annually. A watch the user can't sustain is worse than none — it produces false confidence that someone is watching.

• Cadence must match evidence speed and human capacity. Pricing pages change monthly; statistics

customers churn," the answer is win/loss interviews and discovery, not a patent sweep. The advisor says so.

• The honest off-ramp. Some questions don't need an investigation: if the question is "why do

Application

This interactive skill asks 3 adaptive questions, then offers numbered, context-aware recommendations.

Question 1: What's on your desk?

"What's the situation? Pick the closest, or describe your own:

1. Suspected competitor move — you think someone is building, entering, or repositioning
2. An artifact to build or refresh — TAM/SAM/SOM, battle card, positioning, ICP/personas, pricing analysis
3. A margin or market-structure puzzle — margins eroding, category shifting, entry decision
4. Standing watch setup — you want ongoing coverage, not a one-off answer"

Question 2: Adaptive follow-up

post, a patent, a customer remark? (Whatever you saw is one discipline already flagging; we'll pick the independent channels that could corroborate it.)"

• If 1 (suspected move): "What tipped you off — a job posting, a pricing change, an exec's

company's position in it?"

• If 2 (artifact): "Which artifact, and does a prior version exist to diff against?"
• If 3 (structure puzzle): "Is the question about the whole industry's structure, or one

week, a half-day a month, a day a quarter?"

• If 4 (watch): "Honestly, how much recurring time can you or your team commit — 30 minutes a

Question 3: Constraints

"Two quick constraints: free sources only or is paid tooling available, and any geographic focus? (Region changes which registries and statistics bureaus apply.)"

Then: Recommend

Synthesize and offer 3-5 numbered recommendations, each naming: discipline mix → cadence → executing skill → artifact fed, with one line on when it's the right choice. Always show why the disciplines were chosen (the mapping-table logic), and include the off-ramp when it's honest. Handle single selection, combinations ("1 and 3"), and custom directions per the facilitation protocol.

Routing quick-reference (from the artifact-mapping table):

Close by offering the one-page triage card (template.md) filled in from the session — the route, its "because," and a review date, as a handoff artifact.

Examples

Sample flow (fictional):

Q1 → User: "1 — I think [Competitor A] is building a data-platform play."

Q2 → User: "Their careers page doubled, mostly data engineers."

Q3 → User: "Free sources, US only."

"Your hiring observation means HUMINT has already flagged once — one discipline is an anecdote,

so the job is corroboration from independent channels. Recommendations:

1. TECHINT sweep this week (patents.google.com + their GitHub org + arXiv affiliations, free) —

a paper-or-patent cluster in the same specialty as the hiring surge is the strongest fusion pair

available. Pick this first: highest corroboration value per hour.

2. SIGINT check, 30 minutes (crt.sh for new subdomains, Wayback diff on their product pages) —

catches launch staging if this is further along than hiring suggests.

3. FININT pass at their next earnings call — listen for capex dodges and segment-language

shifts; free via their IR page. Slower, but turns hypothesis into briefable intelligence.

4. Set the watch instead — if this can't get hours this quarter, wire [Competitor A] into

competitive-intel-watch monthly and let the cadence catch it.

Reply 1, 2, 3, 4, a combination like '1 and 2', or tell me more. (Two agreeing disciplines = working

hypothesis; three = brief your leadership.)"

The off-ramp in action: user picks "artifact: ICP refresh," but Q2 reveals the real question is "why did our last three enterprise deals stall?" The honest recommendation leads with win/loss interviews (HUMINT's ground-truth layer) and discovery-interview-prep — "no public-web sweep answers a question your own churned prospects can answer better."

See examples/conversation-flow.md for a full end-to-end session: inline input crediting two of the three questions, a capacity answer that gets believed, a combination selection, and a "tell me more" that earns a teaching answer. examples/conversation-flow-industrial.md shows the routing shift for a physical-world signal — permits and customs data instead of site diffs.

Common Pitfalls

channels matched to the decision beats coverage theater — the mapping table exists so you can skip.

• Prescribing the full eight. Recommending every discipline is refusing to triage. Two or three