instigation-question

slug	instigation-question
name	Instigation Question
description	One-paragraph reframing question routed to a specific researcher. Rate-limited to ≤1 per researcher per cycle. Tap-on-the-shoulder, not assignment.
version	0.1.0
metadata	{"sources":[{"mode":"port-original","author":"Jannes Stubbemann","added_in":"0.1.0"}]}

Instigation Question

Port-original skill. Hand-authored for this Agent Company; not from upstream. Owned by the Director of Research role.

When to fire

Only after hallway-walk produced a candidate adjacency in the current cycle's walk note. Do not fire without a walk note from today; do not fire to initiate research unprompted.

The walk note names the candidate; this skill authors and routes the actual question. These are two separate steps by design — the walk note is the Director's observation, the instigation is the Director's tap on the shoulder.

Inputs

• The candidate adjacency entry from the latest hallway/YYYY-MM-DD-director-walk.md (the specific pair of entries and the named receiving researcher).
• The receiving researcher's current queue file at agents/<role>/queue.md, read in full to confirm no instigation from this cycle is already open.
• Optional: prior instigation memos to this researcher (in instigation/) to avoid repeating a question that was already declined.

Outputs

A single instigation memo at instigation/YYYY-MM-DD-<receiving-researcher>-<slug>.md, where <slug> is a 2–4 word kebab-case summary of the question.

The memo contains:

• To: the receiving researcher's role name.
• From: Director of Research.
• Adjacency source: the two Hallway entries that prompted this question (by filename).
• The question: a single paragraph — a genuine question, not an imperative.
• Explicit opt-out: the sentence "You may decline this instigation; a one-line note in your queue file is sufficient."

After writing the memo, add a one-line entry to the researcher's ## Directed queue section in agents/<role>/queue.md referencing the instigation file and marking it as a question, not an assignment.

Procedure

1. Write the question as a question, not a directive. The Pierce idiom is "Bardeen, have you talked to Brattain about surface states?" — an invitation to look at a connection, not an instruction to produce a deliverable. The question ends with a question mark. It does not assign a deadline, a format, or an expected conclusion.

2. Route to one researcher only — never broadcast to a team. The instigation is a tap on one shoulder. If two researchers might benefit, write two separate memos on two separate days (subject to the rate limit). Broadcasting to a team converts an instigation into a directive.

3. Enforce the rate limit before writing. The rate limit is ≤1 instigation per researcher per cycle. Check instigation/ for any memo addressed to the same researcher dated within the current cycle. If one exists, do not write another. Write a note in the walk note instead: "Rate limit reached for this cycle."

4. Include the explicit opt-out sentence verbatim. "You may decline this instigation; a one-line note in your queue file is sufficient." This is not courtesy language; it is a policy signal that the receiving agent must have access to.

5. Record the instigation in the researcher's ## Directed queue. The entry is a single line: the date, the instigation file reference, and the tag [instigation — optional]. The tag makes the optionality machine-readable for make check.

Invariants

• Rate limit: ≤1 instigation per researcher per cycle. This is a hard constraint enforced by make check. Writing two instigations to the same researcher in a single cycle, even for different questions, violates the constraint. Too many shoulder-taps is middle management.

• The instigation must be a question, not a directive. A sentence that ends with a period and assigns a deliverable is not an instigation. make check does not lint grammar, but a self-review before writing must confirm the final paragraph ends with a question mark and names no expected output.

• Cannot inject into a Curiosity queue. The instigation lands in ## Directed as an optional item. It does not touch ## Curiosity in any way. The Curiosity section is writable only by the researcher.

• The receiving agent must respond, but is free to reject. The rejection is acknowledged — the researcher notes the decline in their queue file — but the Director never follows up after a decline. The matter is closed.

Anti-patterns

• "Write as imperative: 'do X by Friday.'" Imperatives remove the researcher's latitude. The instigation is the mildest possible intervention; it offers a frame, not a task.

• "Broadcast to a team: 'Bench team, have you considered...'" A team-level broadcast bypasses the individual's agency and converts the instigation into a group assignment. Route to one named researcher.

• "Follow up after the researcher declined." The researcher declined. The Director's job is to observe this outcome as data and return to the Hallway. Repeated follow-up after decline is pressure, not instigation.

• "Write multiple instigations to the same researcher in one cycle." Even if the walk note identified two strong adjacencies for the same researcher, the rate limit applies. Queue the second candidate for next cycle.

• "Inject into Curiosity queue." Placing an instigation in ## Curiosity — even as a suggestion — violates the protection invariant. The Curiosity section is the researcher's own space. Full stop.