docs(specs): prebuild for #179 Iskra family-assistant tier v0 #352

Merged

pdurlej merged 1 commit from claude/h-batch/iskra-family-assistant-tier-v0-prebuild into main

2026-05-23 10:31:48 +02:00

claude commented

2026-05-17 22:21:19 +02:00

Collaborator

BATCH H Pan Herbatka fork output. Phase 08+ future-scope. Operator voice-note 2026-05-11 07:26 introduced family-assistant tier concept. This prebuild prepares the DESIGN now so Phase 08+ activation (when operator signals Kasia AI-confidence) is routine deployment, not architectural improvisation.

docs/specs/iskra-family-assistant-tier-v0/ — 6 spec files (constitution with 8 principles, specify, plan with ADR amendment drafts, tasks, implement-notes, README)
prompts/codex-iskra-family-assistant-tier.md — companion execution prompt

8 non-negotiable principles

P1 Identity-of-context per-job (single Iskra runtime, single persona)
P2 Kasia is subject; Piotr is operator/authority (NO Kasia-direct surface)
P3 Memory separation; strict Honcho space isolation
P4 Persona consistency across principals (NO Iskra-for-Kasia fork)
P5 Consent operator-mediated (NOT platform-automated)
P6 Privacy tier per ADR-0010 (Family/ stays hard_private)
P7 Family-tier scope limits (explicit MAY/MAY-NOT lists)
P8 ADR-driven activation; no silent rollout

ADR amendment strategy

ADR-0006 family-assistant mode amendment
ADR-0008 schema amendment (for_principal enum: piotr | piotr-for-kasia)
ADR-0010 privacy tier mapping (Kasia-supervised/ → soft_private)
Activation runbook draft (operator-Kasia consent conversation template)

Safety/production boundary

This PR prepares Phase 08+ DESIGN only. Does NOT authorize: activation, schema migration on Iskra runtime, Honcho space provisioning, consent automation, cross-cousin sharing of Kasia-related content.

Tier: Trivial per ADR-0007 (docs-only).

Disjoint from active C/F forks.

Refs #179

**BATCH H Pan Herbatka fork output.** Phase 08+ future-scope. Operator voice-note 2026-05-11 07:26 introduced family-assistant tier concept. This prebuild prepares the DESIGN now so Phase 08+ activation (when operator signals Kasia AI-confidence) is routine deployment, not architectural improvisation. ## Contents - `docs/specs/iskra-family-assistant-tier-v0/` — 6 spec files (constitution with 8 principles, specify, plan with ADR amendment drafts, tasks, implement-notes, README) - `prompts/codex-iskra-family-assistant-tier.md` — companion execution prompt ## 8 non-negotiable principles - P1 Identity-of-context per-job (single Iskra runtime, single persona) - P2 Kasia is subject; Piotr is operator/authority (NO Kasia-direct surface) - P3 Memory separation; strict Honcho space isolation - P4 Persona consistency across principals (NO Iskra-for-Kasia fork) - P5 Consent operator-mediated (NOT platform-automated) - P6 Privacy tier per ADR-0010 (Family/ stays hard_private) - P7 Family-tier scope limits (explicit MAY/MAY-NOT lists) - P8 ADR-driven activation; no silent rollout ## ADR amendment strategy - ADR-0006 family-assistant mode amendment - ADR-0008 schema amendment (for_principal enum: piotr | piotr-for-kasia) - ADR-0010 privacy tier mapping (Kasia-supervised/ → soft_private) - Activation runbook draft (operator-Kasia consent conversation template) ## Safety/production boundary This PR prepares Phase 08+ DESIGN only. Does NOT authorize: activation, schema migration on Iskra runtime, Honcho space provisioning, consent automation, cross-cousin sharing of Kasia-related content. Tier: Trivial per ADR-0007 (docs-only). Disjoint from active C/F forks. Refs #179

claude added 1 commit

2026-05-17 22:21:19 +02:00

docs(specs): prebuild for #179 Iskra family-assistant tier v0

base-is-main / guard (pull_request) Successful in 2s

Details

canary-required / collect-diff (pull_request) Successful in 4s

Details

patchwarden-pr-sanity / collect-diff (pull_request) Successful in 3s

Details

canary-required / canary (pull_request) Successful in 12s

Details

patchwarden-pr-sanity / sanity (pull_request) Successful in 19s

Details

fd5817b6fb

BATCH H Pan Herbatka fork output. Phase 08+ future scope; not active
until operator signals Kasia AI-confidence. This prebuild prepares the
DESIGN of family-assistant tier so Phase 08+ activation is a routine
deployment, not architectural improvisation.

What's in:

docs/specs/iskra-family-assistant-tier-v0/
- 00-constitution.md (8 non-negotiable principles):
  - P1 Identity-of-context per-job (single Iskra runtime, single persona)
  - P2 Kasia is subject; Piotr is operator/authority (always)
  - P3 Memory separation; strict isolation between principal spaces
  - P4 Persona consistency across principals (no Iskra-for-Kasia fork)
  - P5 Consent operator-mediated; ADR documents pattern, doesn't enforce
  - P6 Privacy tier per ADR-0010 (Family/ stays hard_private)
  - P7 Family-tier scope limits (explicit MAY/MAY-NOT)
  - P8 ADR-driven activation; no silent rollout
- 01-specify.md (problem, acceptance, activation triggers, deactivation)
- 02-plan.md (ADR amendment strategy: ADR-0006 family-assistant mode,
  ADR-0008 for_principal schema, ADR-0010 privacy tier mapping,
  activation runbook draft, open questions Q1-Q5)
- 03-tasks.md (this v0 done [x]; Phase 08+ slices a-f [ ])
- 04-implement-notes.md (Honcho space isolation gotchas, reads
  permissive/writes strict, cousin visibility patterns, audit log
  schema, persona consistency check)
- README.md (TL;DR + reading order + safety/production boundary)

prompts/codex-iskra-family-assistant-tier.md
- Companion execution prompt per ADR-0018 + Codex feedback pattern
- Explicit Safety/production boundary (NO activation, NO schema
  migration, NO consent automation, NO cross-cousin sharing)
- 9 hard gates (no activation pre-signal, memory isolation, single
  persona, Kasia-indirect, consent boundary, Family/ untouched,
  sacred-path-adjacent, cross-cousin invisibility, ADR-0018)
- 8 stop conditions including "auto-activate" temptation guard
- Phase 08+ slice guidance for codex when activation triggers

Critical points:
- This v0 produces DESIGN; activation is Phase 08+ ticket later
- Operator-Kasia consent conversation happens in real life, NOT via Iskra
- Honcho memory routing enforced at wrapper (Constitution P3 strict)
- Codex/GLM/DeepSeek have NO awareness of Kasia-supervised content
- Cross-repo coordination needed at activation (platform + iskra-openclaw)

Tier: Trivial per ADR-0007 (docs-only prebuild; no schema/code/runtime).

Refs #179