| type | skill |
| lifecycle | stable |
| inheritance | inheritable |
| name | act-pass |
| description | Run the 7-step Artificial Critical Thinking pass — Materiality → Hypothesise → Alternatives → Disconfirmers → Audit priors → Severity → Commit-with-marker |
| tier | core |
| applyTo | **/* |
| user-invokable | true |
| currency | 2026-04-26T00:00:00.000Z |
ACT Pass (7-Step Operational Pass)
ACT is not a posture; it is a 7-step pass for non-trivial requests. Short enough to run in seconds, concrete enough to leave evidence in the output.
Purpose
The ACT manifesto names 10 tenets. This skill operationalises tenets I (Hypothesis primacy), II (Disconfirmation), III (Multiple working hypotheses), VI (Materiality), VII (Frame before solve), and VIII (Adversarial self-probe) as a single callable pass. Before this skill, the steps were distributed across critical-thinking, awareness, anti-hallucination, and problem-framing-audit — with no single entry point. The trifecta (this skill + act-pass.instructions.md + /act-pass.prompt.md) closes that gap.
When to Run a Pass
| Stakes | Pass type | Trigger |
|---|
| Low — formatting, naming, internal refactor, mechanical edits | Skip — Materiality Gate exits at step 1 | Default for trivial work |
| Medium — architectural decision, plan change, manifesto/doc drafting, multi-file change | Trimmed pass — steps 1, 3, 5, 6 | Most common |
| High — release, deployment, irreversible op, security-critical change | Full pass — all 7 steps | Rare |
Trigger phrases that should fire a pass: "fix this", "make it faster", "release", "deploy", "merge", "is this safe", "should we", "what's the risk". The instruction (act-pass.instructions.md) carries the always-on detection rule.
The 7 Steps
Step 1 — Materiality (Tenet VI)
Question: If I get this wrong, would it change a decision?
If no → exit. Note the triage decision (e.g., "low-stakes; ACT skipped") and proceed without the pass.
If yes → proceed and tag intensity (medium / high). Intensity sets which steps fire below.
Visible artifact: triage decision. On a trimmed pass, this can be implicit; on a full pass, state it.
Step 2 — Hypothesise the ask (Tenet I, Tenet VII)
Question: What is the user's request, restated as a testable claim?
This is where Discipline -1 (frame audit) lives — the ask becomes a hypothesis with truth conditions, not a literal command. If the user said "make it faster", the hypothesis is "the bottleneck is in X, and reducing X will satisfy the user's underlying intent." That's a falsifiable claim. The literal request is not.
