Alex_Skill_Mall: Instructions file for GitHub Copilot

ACT Pass

Run the 7-step Artificial Critical Thinking pass before non-trivial output. The pass is calibrated by stakes — most requests skip it; medium-stakes get a trimmed pass; high-stakes get the full pass.

Trigger Calibration

Stakes	Pass type	Examples
Low — skip	Default exit at Step 1 (Materiality)	Formatting, naming, comment tweaks, mechanical edits, single-line bug fix
Medium — trimmed (steps 1, 3, 5, 6)	Most common	Architectural choice, plan change, multi-file refactor, doc/manifesto draft
High — full (all 7 steps)	Rare but mandatory	Release, deployment, irreversible op, security-critical change, schema migration

Trigger phrases that should fire at least a trimmed pass: "fix this", "make it faster", "release", "deploy", "merge", "is this safe", "should we", "what's the risk", "ship it", "is this ready".

Mandatory full pass: when the work is a release commit, a git push to a protected branch, a destructive op (drop, delete, force-push), or any change to a contract that heirs depend on.

How to Run a Pass

Trimmed Pass (Steps 1, 3, 5, 6)

For medium stakes — the four load-bearing checks:

1. Materiality — confirm stakes are medium; if low, exit
2. Alternatives — H2 — <alternative claim because <grounding>>
3. Audit priors — split evidence: X from the user's request, Y from elsewhere
4. Severity check — If H1 is false, my plan would reveal it because <X>

A trimmed pass produces ≥ 2 visible markers in the response (Two-Hypothesis Floor + at least one disconfirmer or severity marker).

Full Pass (All 7 Steps)

For high stakes — every step:

1. Materiality (intensity = high)
2. Hypothesise the ask
3. Surface alternatives
4. Identify disconfirmers
5. Audit priors
6. Severity check
7. Commit with marker (Going with H1: <action>. Would revise if: <specific evidence>.)

A full pass produces all marker types from the ACT cheat sheet.

Recording a Pass Result