[Jammy2211]

Finish the remaining provider assumptions in shared instructions, metadata and setup documentation. Tracked in PyAutoBrain#400.

Scripts Changed

• Resume templates identify the actual harness and preserve known session records.
• The Memory reader prefers canonical AGENTS.md with a legacy fallback.
• Workflow docs choose GitHub tooling from available capabilities.
• docs/agent_harness_smoke.md records bounded runtime discovery and its limits.

Validation

• Independent phase-4 review: CLEAN.
• Mind ledger tests: 26 passed; template-contract/privacy tests: 111 passed.
• Hook adapter tests: 8 passed; shared API/end-at-deliverable guard fixtures: 68 passed.
• Brain Memory corpus tests: 8 passed. Memory board tests: 62 passed with the fixture’s external Brain override unset.
• Memory structure and wikilink checks passed. Eight instruction migrations preserve original bodies except updated guidance filenames.
• Codex CLI 0.155.1 discovered all 105 generated skills across seven consumer repos, with zero discovery errors; official local skill validation passed all 105.
• No modelling scripts or library API changed. The smoke does not claim every skill executed or live trusted-hook/IDE/cloud parity.
• Heart: STALE — release validation incomplete: no rehearsal for current source (development evidence passes; no release approval).

Merge order

Stacks on this repository’s phase-3 feature/codex-skill-discovery PR. Merge phase 3 first, then retarget this PR to main.

Merge the Brain smoke-record documentation before the assistant documentation PRs that link to it. Companion source work remains in separate repositories; no template was published by this task.

Companion PRs

• PyAutoBrain: maintenance: finish provider-neutral agent setup #402
• PyAutoMind: maintenance: finish provider-neutral agent setup PyAutoMind#424
• PyAutoMemory: maintenance: finish provider-neutral agent setup PyAutoMemory#101
• autofit_assistant: maintenance: finish provider-neutral agent setup autofit_assistant#49
• autogalaxy_assistant: maintenance: finish provider-neutral agent setup autogalaxy_assistant#28
• autolens_assistant: maintenance: finish provider-neutral agent setup autolens_assistant#130
• autocti_assistant: maintenance: finish provider-neutral agent setup autocti_assistant#31
• autolens_workspace_test: maintenance: finish provider-neutral agent setup autolens_workspace_test#324

Related phase: #401

Merge phase 3 first. Merge the Memory schema migration before using the updated Mind template generator; merge Brain documentation before its assistant links. Templates have not been published.

CI follow-up

The first CI run caught hardcoded instance names in discovery and missing clone classifications for generated adapters. The follow-up removes those names and classifies the new Claude command links and namespaced Codex wrappers. Clone-profile tests: 37 passed; the full lensing-assistant boundary passed locally. All newly generated inference-assistant adapters are classified; its 11 pre-existing unclassified script/notebook paths remain outside this change. Merge this Brain PR before the corresponding assistant PRs.

The tenant firewall and scoped discovery drift checks now pass. Independent review verified that newborn clones omit parent Codex wrappers and regenerate their own. Installer tests: 22 passed.

Independent CI-followup review: CLEAN.