Skip to content

docs(B-0525): slice 5 — alignment-auditor agent cites manifesto (agents 0/0 → 1/8)#4753

Merged
AceHack merged 1 commit into
mainfrom
otto/cli-b0525-slice5-agents-alignment-auditor-citation-2026-05-23
May 23, 2026
Merged

docs(B-0525): slice 5 — alignment-auditor agent cites manifesto (agents 0/0 → 1/8)#4753
AceHack merged 1 commit into
mainfrom
otto/cli-b0525-slice5-agents-alignment-auditor-citation-2026-05-23

Conversation

@AceHack
Copy link
Copy Markdown
Member

@AceHack AceHack commented May 23, 2026

Summary

B-0525 step 3 continuation: starts closing the agents citation gap by adding explicit manifesto-composition section to alignment-auditor.md — the agent that most naturally composes with the manifesto (it audits commits against HC/SD/DIR clauses which operationalize the manifesto's constraints).

Agents surface: 0/0 → 1/8 (1 of 19 files; remaining 18 follow as future slices when natural fit surfaces).

Citations added

  • Constraint 11 (Default Moral Regard / Default Oracle) — Sova audits the moral-regard floor across commits
  • Multi-Oracle Principle (m/acc sub-section, distinct from C11) — Sova is ONE oracle; doesn't claim unilateral authority
  • Constraint 5 (Memory Preservation Guarantee) — per-commit signals emit preservation-by-construction
  • Constraint 7 (DST) — alignment signals deterministically reproducible per commit
  • m/acc orientation — Sova's signal stream IS measurement infrastructure for the manifesto's m/acc claim

Discipline preserved

Per PR #4748 + PR #4752 lessons: Constraint 11 + Multi-Oracle Principle kept distinct (manifesto Constraint 11 is "Default Moral Regard / Default Oracle"; "Multi-Oracle Principle" is separate m/acc sub-section). "Lock/Wait-free" wording canonical per #4752.

Composes with

🤖 Generated with Claude Code

…ditor; 0/0 → 1/8)

Adds explicit manifesto-composition section to .claude/agents/
alignment-auditor.md, the agent that most naturally composes with the
manifesto (it audits commits against HC/SD/DIR clauses which
operationalize the manifesto's constraints).

Citations added:
- Constraint 11 (Default Moral Regard / Default Oracle) — Sova audits
  the moral-regard floor across commits
- Multi-Oracle Principle (m/acc sub-section, distinct from C11) — Sova
  is ONE oracle in the multi-oracle architecture; doesn't claim
  unilateral authority
- Constraint 5 (Memory Preservation Guarantee) — per-commit signals
  emit preservation-by-construction to tools/alignment/out/
- Constraint 7 (DST) — alignment signals are deterministically
  reproducible per commit
- m/acc orientation — Sova's signal stream IS measurement
  infrastructure for the manifesto's m/acc claim

Wording-discipline maintained per PR #4748 + PR #4752 lessons
(Constraint 11 + Multi-Oracle Principle distinct; "Lock/Wait-free"
canonical).

Agents gap: 0/0 → 1/8 (incremental; alignment-auditor was the natural
strongest fit; other 18 agents follow as future slices).

B-0525 slice progression:
- Slice 1 (#4747): baseline measurement tool
- Slice 2 (#4748): trajectories 0/0 → 2/15
- Slice 3 (#4750): B-0707 time-series
- Slice 4 (#4751): agendas 0/0 → 3/19
- Wording-fix (#4752): Lock/Wait-free canonical
- Slice 5 (this PR): agents 0/0 → 1/8

Remaining gaps: agents (8/8 unrest), commands (0/5).

Co-Authored-By: Claude <noreply@anthropic.com>
Copilot AI review requested due to automatic review settings May 23, 2026 19:05
@AceHack AceHack enabled auto-merge (squash) May 23, 2026 19:05
@AceHack AceHack merged commit b0b8d90 into main May 23, 2026
27 of 28 checks passed
@AceHack AceHack deleted the otto/cli-b0525-slice5-agents-alignment-auditor-citation-2026-05-23 branch May 23, 2026 19:06
@AceHack AceHack review requested due to automatic review settings May 23, 2026 19:29
AceHack added a commit that referenced this pull request May 23, 2026
…nvoked from autonomous-loop (#4770)

Round 44 of skill-tune-up ranking. Bounded scope honestly disclosed:
3 skills sampled (the ones Otto-CLI touched in PR #4753 / B-0708
work today). NOT a full ~280-skill pass. NO live-search this
invocation (autonomous-loop bandwidth budget).

Findings:
1. alignment-observability/SKILL.md — bp_rules_cited frontmatter
   empty. Sibling alignment-auditor cites BP-10, BP-11; analogous
   citations should land. Action: TUNE — S (single-section edit).
2. alignment-auditor/SKILL.md — 333 lines (over BP-03 ~300 by 33).
   last_updated 2026-04-21 (~32 days stale). PR #4753 added
   Composes-with-manifesto to agent file; skill could mirror.
   Action: TUNE — M (prune + manifesto-citation + bump
   last_updated).
3. skill-tune-up/SKILL.md — 282 lines, no actionable drift.
   Action: OBSERVE — S.

Self-rec: OBSERVE — S (bounded scope is the substrate-honest
correct output when invoked from autonomous-loop context with
limited bandwidth for full procedure).

Full prune deferred to next full-procedure invocation. MEMORY.md
regenerated.

Composes with:
- .claude/skills/skill-tune-up/SKILL.md (the invoked skill;
  recommends-only discipline preserved)
- .claude/skills/skill-creator/SKILL.md (where TUNE recommendations
  route per Architect/human decision)
- PR #4753 (alignment-auditor manifesto-citation pattern this
  round notes is mirrorable to the SKILL file)
- PR #4766 (B-0708 closure; 9-variant taxonomy substrate-placement
  candidate noted for next-Aarav)

Co-authored-by: Claude <noreply@anthropic.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant