# Immune-Brain Host Preferences Pi and Claude Code are supported code-agent hosts. Immune-Brain does not load an agent-local TOML file or Immune-Brain-specific environment overrides. User and project preferences belong in the Host's agent instruction files. ## Precedence Planner preferences resolve in this order: 1. a literal instruction in the current request; 2. the repository root agent instruction file, whichever the repository tracks: `AGENTS.md` or `CLAUDE.md`; 3. the Host's user-level agent instruction file; or 4. an explicit user question. Hosts differ in which files they auto-load: Pi injects `AGENTS.md`, while Claude Code auto-loads `CLAUDE.md` and does not read `~/.pi/agent/AGENTS.md` at all. A Skill therefore reads sources 2 and 3 directly instead of assuming the Host placed them in context, and reports which sources it checked. Invalid values are reported rather than guessed. A preference with no documented default resolves to a user question, never to a silently chosen value. ## Initiative Carrier The Initiative carrier preference applies only to proposals split across multiple TaskIntents. Ordinary TaskIntents remain tracked by Kernel TaskRecords. There is no built-in carrier default; when no directive is found, Planner asks. Set one of these fixed directives in the repository root `AGENTS.md` or `CLAUDE.md`: ```md ## Immune-Brain Preferences - Initiative carrier default: local ``` ```md ## Immune-Brain Preferences - Initiative carrier default: github ``` A repository directive overrides the user-level directive. A configured `github` default is standing opt-in for GitHub projection, but the literal user still confirms the Initiative name, immutable slug, complete Parent/Child decomposition, granularity, and dependencies before the first remote mutation. Planner reports the selected carrier and its source. After confirmation it publishes the complete Issue graph in one idempotent batch and reports the recommended first unblocked Task, stable dependency order, and parallel groups. Projection failure is reported with a batch retry action and does not invalidate the authored planning files, but it blocks Enrollment and execution handoff for that Initiative until the same complete batch succeeds. It never silently switches carrier. ## Lane Preferences Four optional directives hold standing lane-mode choices. The Parent resolves them before the first `start_unattended_batch` call for an Initiative; the batch runner reads none of them. None has a built-in default. ```md ## Immune-Brain Preferences - Lane max parallel: 4 - Lane Executor Host: claude-code - Lane Executor model: - Lane Executor effort: high ``` - `Lane max parallel` is passed as `max_parallel` and turns lane mode on. Without it a batch stays serial unless you ask for lanes in that request. A resumed batch keeps the value it recorded. - `Lane Executor Host` is `claude-code` or `pi`. Every Lane uses it, with no fallback to the other Host. Without it the Parent prefers its own Host type. - `Lane Executor model` and `Lane Executor effort` apply only when `Lane Executor Host` is set, because a model ID belongs to one Host. Without them each Executor Host uses its own default. Lane-mode agents only create: they close no tab, stop no session and remove no Lane. The full contract is `plugins/immune-brain/dist/drive-batch.md#lane-preferences`. ## Risk Floor The paths that force a higher task risk are declared by the project in one tracked file, `.imm/risk-floor.json`. It is not a preference an Agent reads: the Kernel enforces it, so it does not live in `AGENTS.md` or `CLAUDE.md`. ```json { "contract": "assurance_kernel/risk_floor/v1", "material": { "paths": ["src/auth"], "changed_only": ["db/migrations"] }, "critical": { "paths": ["infra/prod"], "changed_only": [] } } ``` - `paths` raises a task whose `scope_hint` can reach the prefix, including through a glob, or whose changed paths sit beneath it. - `changed_only` raises a task only through its changed paths, so a scope that merely names the prefix keeps its declared risk. - Every entry is a literal, normalized, project-relative path prefix with forward slashes and no wildcard. Both tiers and both lists are optional. - A declared risk is only ever raised, never lowered. The reader fails closed: - No file means no floor. Every task keeps the risk its TaskIntent declares. - A file that is not valid JSON, has an unknown field or a wrong type, is not a regular file, exceeds 64 KiB, or cannot be read makes every task of the project `critical`. - A file with staged, unstaged or untracked changes is treated the same way until it is committed, so an edit made beside a task cannot lower that task's floor. - `.imm/risk-floor.json` itself is always `critical`: a task whose scope can reach it, or that changes it, is `critical` whatever the file says. `imm-kernel inspect --json` reports the prefixes in force as `risk.floor_prefixes` and the reason an untrusted declaration was refused as `risk.floor_invalid`. A TaskRecord keeps the risk it was enrolled with; changing the declaration does not re-grade a task that is already enrolled. ADR 0023 records the decision. ## Git Integration One line in the tracked repository root `AGENTS.md`, else in the tracked root `CLAUDE.md`, decides whether the runtime writes commits to your branch on its own: ```markdown - Git integration: manual ``` - `manual` is the default, used when the line is absent. The runtime writes no commit outside batches. Enrollment stages its TaskIntent and Spec instead of committing them; with an unborn HEAD, in-scope changes are claimed as task work; storage-layout migration leaves its diff in the working tree and lists the paths. A settled task is delivered without any commit: commit however you prefer. - `auto` keeps the precondition commits of ADR 0023 and, when a task settles `done`, commits exactly its settled scoped paths as `imm(): settled delivery`. - Any other value behaves as `manual`. The status Tool reports it as an `invalid_git_integration` warning naming the file and the value. - The status Tool reports the resolved preference as `git_integration`. - The confirmed empty root commit for an unborn HEAD is made in both modes. ADR 0024 records the decision. ## Other Preferences - Set reply-language preferences as ordinary `AGENTS.md` communication instructions. Machine contracts, schema fields, paths, API names, and code identifiers remain literal. - Project `AGENTS.md` owns standing authorization for bounded advisory subagents. Explicit solo instructions and Pi host policy still take precedence. - Agent model selection uses the active Pi session model unless a Pi `Agent` invocation explicitly selects another Pi-configured model.