Autonomy-level advisor

Decide the maximum AI autonomy level (L0–L5, as in this chapter) allowed for proposed tasks.

Input: lines task risk reversibility data checks until the end of input:

• risk: LOW, MEDIUM, HIGH (impact on people, security, money)
• reversibility: EASY (rollback in minutes) or HARD
• data: NONE, PERSONAL, SENSITIVE
• checks: WEAK or STRONG (quality of automated tests/evals/scans for this change)

Rules — start at L4 and lower the level:

1. risk HIGH → at most L2; risk MEDIUM → at most L3.
2. reversibility HARD → lower by 1.
3. data PERSONAL → at most L2; data SENSITIVE → at most L1.
4. checks WEAK → lower by 1.
5. The level never goes below L0. (L5 is never allowed.)

Apply rules 1 and 3 as caps (minimum with the current level) and rules 2 and 4 as reductions, in the order 1, 2, 3, 4.

Output per task: task: L<n> - <control> with the control for the resulting level:

• L4: auto-merge if gates pass; audit samples
• L3: agent PR; human reviews every change
• L2: AI assists on request; human writes and commits
• L1: suggestions only
• L0: human only

Then Tasks: n, human-reviewed (L0-L3): h.

Input:

readme-typos LOW EASY NONE STRONG
search-filter MEDIUM EASY NONE STRONG
hidden-post-rule HIGH HARD PERSONAL STRONG
harassment-reply HIGH HARD SENSITIVE WEAK

Output:

readme-typos: L4 - auto-merge if gates pass; audit samples
search-filter: L3 - agent PR; human reviews every change
hidden-post-rule: L1 - suggestions only
harassment-reply: L0 - human only
Tasks: 4, human-reviewed (L0-L3): 3