llm-wb 0.1.0-beta.0

package/docs/00.chat/llm-workbench-acceptance-matrix.md

@@ -0,0 +1,55 @@
+<!-- agentic-artifact:
+  schema: agentic-artifact/v2
+  id: chat.docs.llm-workbench-acceptance-matrix
+  version: 1
+  status: active
+  layer: 00.chat
+  domain: portability
+  disciplines:
+  - agentic
+  kind: doc
+  purpose: Map llm-workbench public-beta invariants to enforcing artifacts and checks.
+  portability:
+    class: required
+    targets:
+    - llm-workbench
+  used_by:
+  - id: chat.standards.llm-workbench-public-beta-contract
+    path: .agentic/00.chat/standards/llm-workbench-public-beta-contract.md
+  - id: chat.script.upstream.validate-llm-workbench-portability.readme
+    path: scripts/00.chat/upstream/validate-llm-workbench-portability/README.md
+-->
+# llm-workbench Acceptance Matrix
+
+This matrix maps the public-beta contract to the artifacts that enforce it.
+
+| Invariant | Primary Artifacts | Required Proof |
+| --- | --- | --- |
+| No durable chat classification | `chat-start.md`, `start-chat-session`, portability validator | Startup metadata has no whole-chat `layer`, `mode`, or `workflow` fields. |
+| Standalone public export | Bootstrap planner, artifact metadata checker, portability validator | Generated public repo and installed target omit source-only trees/scripts and pass `check-headers --all`. |
+| No source-specific generic gates | `prepare-chat-session-before-commit`, before-commit checklist, portability validator | Generic chat surfaces do not hard-code RAG/rulebook gates; optional gates use neutral hooks. |
+| Provider-neutral transcript metrics | `record-chat-commit`, transcript scripts, startup templates | Missing transcript metrics are portable by default; Codex discovery is opt-in adapter behavior. |
+| Provider-neutral cost metrics | `estimate-chat-cost`, `scripts/00.chat/metrics/data/`, record-commit smoke test | Default pricing is `portable-unpriced`; explicit profile overrides record cost. |
+| Safe install/uninstall | Public install/uninstall templates, install smoke, portability validator | Dry run is read-only, apply preserves user files, uninstall removes only manifest-owned material. |
+| Thin assistant adapters | Public adapter templates, install manifest, portability validator | Adapters route to chat-start and avoid provider-specific policy duplication. |
+| Public CLI wrapper | `bin/llm-workbench.js`, package `bin`, package `files`, CLI smoke | `npm pack --dry-run` includes the CLI and required scripts; `init`, `list`, `sessions list`, `new`, `commit`, and `merge-main` are smoke-tested. |
+| CLI/code-assistant robustness | Dispatcher, startup JSON mode, portability validator | `llm-wb new --json` works; stripped executable bits do not break dispatch. |
+| Temporary eval loop | Public-beta checklist, closeout evidence | Temporary evals fail first, pass after implementation, and are removed before closeout. |
+| Documentation truthfulness | Public README/docs/templates | Public docs use the accepted standalone/provider-neutral claim and avoid universal-assistant overclaims. |
+
+The permanent acceptance entrypoint is:
+
+```bash
+bash scripts/00.chat/upstream/validate-llm-workbench-portability/script.sh
+```
+
+Changes that discover a new edge case should add that edge case to the
+portability validator or a focused smoke test before closeout.
+
+Source-only public-export exclusions include `.agentic/agentic`, `.docs`,
+`.scripts`, `.agentic/01.harness`, `.agentic/docs`, `.agentic/scripts`, and
+`scripts/00.chat/classification`, plus maintainer-history ADR surfaces such as
+`docs/00.chat/public-chat-workbench-adrs.md` and
+`docs/harness/architecture/adrs/`. Public `scripts/01.harness` is limited to
+portable validation and governed-runner helpers; source-maintenance migration,
+taxonomy-generation, and rule-test scripts must not be exported or installed.

package/docs/00.chat/script-layout.md

@@ -0,0 +1,107 @@
+<!-- agentic-artifact:
+  schema: agentic-artifact/v2
+  id: chat.script-layout
+  version: 1
+  status: active
+  layer: 00.chat
+  domain: scripts
+  disciplines:
+  - agentic
+  kind: doc
+  purpose: Explain the current script layout after the chat harness script migration.
+  portability:
+    class: required
+    targets:
+    - llm-workbench
+  used_by:
+  - id: harness.readme
+  - id: harness.architecture.adr.0017-organize-scripts-by-owner-domain-and-capability
+  - id: harness.architecture.adr.0020-use-scripts-for-layer-command-surfaces
+-->
+# Script Layout
+
+This document describes the current script layout. Public workbench users
+should treat this file and the current scripts as the operational source of
+truth; maintainer ADRs remain source-side history only.
+
+## Current Shape
+
+Chat lifecycle scripts live under:
+
+```txt
+scripts/00.chat/<domain>/<capability>/
+```
+
+Shared harness governance scripts live under:
+
+```txt
+scripts/01.harness/
+```
+
+Future RAG/rulebook, product, and deployment command surfaces should live under:
+
+```txt
+scripts/02.rag-rulebook/<domain>/<capability>/
+scripts/03.product/<domain>/<capability>/
+scripts/04.deploy/<domain>/<capability>/
+```
+
+That split is intentional:
+
+- `scripts/00.chat/` owns chat startup, commands, reporting, session logs,
+  recovery, worktrees, main refresh, local merge checks, and upstream bootstrap
+  support.
+- `scripts/01.harness/` owns cross-layer harness checks such as metadata,
+  deterministic process drift, governed command drift, and the governed script
+  runner.
+- `scripts/02.rag-rulebook/` is reserved for future standalone RAG/rulebook
+  automation such as corpus extraction, rulebook index generation, chunk
+  generation, graph expansion, and context-packet validation.
+- `scripts/03.product/` is reserved for future product-owned automation such as
+  entity-builder commands, code generation, migration helpers, product
+  validation, and developer CLI capabilities.
+- `scripts/04.deploy/` is reserved for future deployment-owned automation
+  such as environment checks, release helpers, deployment validation, and
+  operational command wrappers.
+
+Use `scripts/` as the canonical executable command surface. The term `tools`
+describes automation capabilities, not a separate top-level command namespace.
+Future MCP exposure should wrap stable script capabilities through an explicit
+registry or manifest instead of importing layer internals directly.
+
+Do not add new chat lifecycle scripts under `scripts/shared/chat/` or
+`scripts/shared/git/`. Those were compatibility locations from the earlier
+layout and are now retired.
+
+## Capability Folders
+
+Each capability folder should contain the files that help a reader understand,
+run, and test that capability:
+
+- `script.sh`, `script.js`, or `lib.sh` for the implementation
+- `smoke-test.sh` or another focused validation file when behavior needs proof
+- `README.md` explaining the purpose, inputs, effects, and boundaries
+
+Domain folders also have a `README.md` that explains how the capabilities in
+that domain fit together.
+
+## Public Surface
+
+Humans should usually use `package.json` `chat:*` scripts. Those package
+scripts delegate to canonical `scripts/00.chat/...` paths.
+
+Harness workflows, checklists, standards, and gates should point directly at
+canonical `scripts/<numbered-layer>/...` paths for the layer that owns the
+behavior.
+
+## Historical Paths
+
+Historical documents may mention retired paths, but should do so as migration
+history:
+
+- say what previously lived there
+- name the current canonical path or say the behavior was retired
+- avoid runnable examples that make the old path look current
+
+Negative tests may create retired paths in throwaway repositories when they are
+proving the governed runner rejects old compatibility surfaces.

package/docs/adapting-to-your-repo.md

@@ -0,0 +1,29 @@
+# Adapting To Your Repo
+
+The workbench is intentionally small at the root and detailed in the harness
+folders.
+
+## Keep AGENTS.md Small
+
+Use `AGENTS.md` as a router. Put detailed rules in workflows, standards,
+checklists, and scripts.
+
+## Add A Layer Deliberately
+
+If your repo needs product, infrastructure, docs, or education governance, add
+a layer under `.agentic/` and update `AGENTS.md` only with the routing summary.
+
+## Prefer Scripts For Repeatable Checks
+
+If a process can be checked deterministically, add or update a governed script
+instead of relying only on prose.
+
+## Keep Session History Local To The Repo
+
+Do not import another repo's `commitLogs/`. The first chat in your repo creates
+your repo's own session history inside the chat-owned worktree.
+
+## Promote Reusable Lessons Upstream
+
+When a target repo teaches you something reusable about the workbench, promote
+that lesson back to `llm-workbench` instead of letting each repo drift.

package/docs/concepts.md

@@ -0,0 +1,38 @@
+# Concepts
+
+## Chat Session
+
+A chat session is one unit of agent-assisted work. The session log records the
+task, branch, workflow, commits, and estimated transcript metrics.
+
+Session logs live under:
+
+```txt
+commitLogs/<year>/<month>/<day>/<session>/README.md
+```
+
+## Chat-Owned Worktree
+
+Implementation work happens in a chat-owned Git worktree. The root repo stays
+available as the local integration console.
+
+This lets a chat branch do real work without blocking other chats or mixing
+unrelated changes.
+
+## Governed Script
+
+A governed script is a helper command that has a known path, metadata, and
+approval boundary. The harness prefers governed scripts over ad hoc shell
+commands when behavior should be repeatable.
+
+## Local Main Refresh
+
+Chat branches should be refreshed from local `main` before promotion. The
+harness separates rehearsal from apply so conflicts can be inspected before
+the branch is changed.
+
+## Commit Log
+
+The commit log is the work ledger for a chat. It is not a changelog for the
+whole project. It records what happened in one session and preserves the
+evidence needed for closeout, reporting, and future recovery.

package/docs/install.md

@@ -0,0 +1,114 @@
+# Install
+
+## Requirements
+
+- Git
+- Bash
+- Node.js and npm for the `llm-wb` CLI
+
+## Install Into A Target Repo
+
+The intended public install path starts from the target Git repo:
+
+```bash
+cd /path/to/target/repo
+
+npx llm-wb init --dry-run
+npx llm-wb init
+```
+
+Dry-run mode prints the package and file plan without writing. Apply mode
+refuses to write when the plan contains conflicts.
+
+The installer copies the workbench harness into the target Git repo. It does not
+copy `commitLogs/` from the workbench repo, overwrite existing assistant
+instruction files, or replace unrelated `package.json` scripts. Existing
+instruction files receive a managed `llm-workbench` block; package files are
+merged instead of overwritten.
+
+For an empty Git repo with no first commit yet:
+
+```bash
+npx llm-wb init --init-commit
+```
+
+That creates an install commit after the clean plan is applied.
+`--init-commit` is only for repos with no existing `HEAD`; existing repos should
+review and commit the install changes through their normal process. The install
+commit stages only workbench-owned files, managed instruction blocks,
+`package.json` when workbench scripts were added, and the install manifest.
+
+Target installs do not copy the upstream `.agentic/01.harness` maintenance
+tree. Reusable harness lessons should be promoted back to `llm-workbench`
+instead of creating source-specific harness governance in each target repo.
+
+## Verify The Install
+
+From the target repo:
+
+```bash
+npx llm-wb list
+npx llm-wb sessions list
+```
+
+`llm-wb list` lists available installed workbench commands. `llm-wb sessions
+list` lists active chat sessions and branches. Both commands should run without
+requiring a current chat session.
+
+The public-beta contract is summarized in `docs/public-beta-contract.md`.
+
+## First Chat
+
+Open the target repo with your coding agent. The first chat startup should
+create that repo's own `commitLogs/` tree inside the chat-owned worktree and
+use the target repo as the source of truth.
+
+For tools that prefer machine-readable startup packets:
+
+```bash
+npx llm-wb new --json "Describe the prompt"
+```
+
+The JSON packet includes the session log, chat-owned worktree, lifecycle
+workflow, latest context-packet references, and first prompt.
+
+## Transcript Metadata
+
+The workbench stores neutral transcript metadata:
+
+- `transcript_provider`
+- `transcript_path`
+- `transcript_bytes`
+- `transcript_source`
+
+Codex transcript discovery can populate these fields automatically. Other
+providers can supply them when recording a commit:
+
+```bash
+CHAT_TRANSCRIPT_PROVIDER=mistral \
+CHAT_TRANSCRIPT_BYTES=4096 \
+CHAT_TRANSCRIPT_SOURCE="Mistral CLI transcript bytes" \
+  bash scripts/00.chat/session-log/record-chat-commit/script.sh \
+    <commit-sha> \
+    "Commit subject" \
+    "Commit summary"
+```
+
+Missing transcript metrics do not block portable mode. Set
+`CHAT_TRANSCRIPT_METRICS_MODE=strict` when a repo wants commit recording to fail
+without transcript metrics.
+
+## Uninstall
+
+From the target repo:
+
+```bash
+bash scripts/uninstall.sh --dry-run .
+bash scripts/uninstall.sh --apply .
+```
+
+Uninstall reads `.llm-workbench/install-manifest.tsv` from the target repo. It
+removes only workbench-owned files, package scripts, and managed instruction
+blocks recorded in that manifest. It should not remove target repo product code,
+unrelated package scripts, or committed session history unless you explicitly
+clean those files yourself.

package/docs/public-beta-contract.md

@@ -0,0 +1,45 @@
+# Public-Beta Contract
+
+`llm-workbench` is a standalone, provider-neutral Git/Bash/npm chat harness for
+blank or existing repos, with safe install/uninstall, assistant adapters, JSON
+startup packets, and portable transcript metadata across common CLI and
+code-assistant workflows.
+
+It does not claim to work perfectly with every LLM assistant or every editor
+setup. Repos can provide instruction files, command surfaces, and structured
+handoffs, but each assistant decides how well it follows those instructions.
+
+## Invariants
+
+- No durable whole-chat `layer`, `mode`, or `workflow` classification.
+- Prompt-level routing is local and optional.
+- Base startup does not require RAG/rulebook, classifier scripts, or
+  source-specific routers.
+- Public export and target installs are self-contained.
+- Transcript metadata is provider-neutral.
+- Codex transcript discovery is an optional adapter, not a default assumption.
+- Cost metrics are unavailable by default until a pricing profile is selected.
+- Install and uninstall are manifest-backed and preserve user files.
+- Assistant adapters are thin routers to the canonical chat lifecycle workflow.
+- The public `llm-wb` CLI is a thin wrapper around the existing
+  installer, dispatcher, session-log, and local-merge scripts.
+- `llm-wb list` lists installed workbench commands. Active chat sessions are
+  listed with `llm-wb sessions list`.
+- `llm-wb new --json` supports code-assistant startup handoff.
+
+## Required Checks
+
+Before changing public harness behavior, run the portability suite:
+
+```bash
+bash scripts/00.chat/upstream/validate-llm-workbench-portability/script.sh
+```
+
+Public changes should also run the relevant focused smoke tests listed in the
+source repo standard.
+
+CLI-facing changes should run:
+
+```bash
+bash tests/smoke-test-cli.sh
+```

package/docs/workflows.md

@@ -0,0 +1,103 @@
+# Workflows
+
+## Start A Chat
+
+Chat startup is governed by:
+
+```txt
+.agentic/00.chat/workflows/chat-start.md
+```
+
+At startup the harness finds or creates session metadata, records the chat
+lifecycle workflow, and verifies the chat-owned worktree when writes are
+allowed. It does not assign the whole chat a durable layer, mode, or workflow.
+
+For assistants that can consume structured startup output:
+
+```bash
+llm-wb new --json "Describe the prompt"
+```
+
+The JSON packet includes the session log, chat-owned worktree, lifecycle
+workflow, latest context-packet references, and first prompt.
+
+Use `npx llm-wb ...` instead of `llm-wb ...` when the CLI is not installed
+globally or linked into your shell.
+
+## Commit Work
+
+Commit preparation is governed by:
+
+```txt
+.agentic/00.chat/workflows/chat-commit.md
+```
+
+The before-commit gates check prerequisites, session log safety, deterministic
+process drift, metadata headers, and governed command drift.
+
+The public CLI shortcut is:
+
+```bash
+llm-wb commit -m "Describe the completed work"
+```
+
+This wraps the existing commit gates, creates the task commit, records it in the
+session log, and checkpoints session evidence.
+
+Transcript metrics are provider-neutral. Codex can use the bundled transcript
+adapter; other assistants can provide values with `CHAT_TRANSCRIPT_PROVIDER`,
+`CHAT_TRANSCRIPT_PATH`, `CHAT_TRANSCRIPT_BYTES`, and `CHAT_TRANSCRIPT_SOURCE`.
+Portable mode records unavailable metrics without blocking. Strict mode is
+available through `CHAT_TRANSCRIPT_METRICS_MODE=strict`.
+
+Cost metrics are provider-neutral by default. Bundled pricing records cost as
+unavailable until `CHAT_COST_PROFILE` or `CHAT_COST_PRICING_FILE` selects a
+concrete pricing profile.
+
+## Refresh From Main
+
+Main refresh is governed by:
+
+```txt
+.agentic/00.chat/workflows/chat-refresh-from-main.md
+```
+
+The workflow separates read-only status, rehearsal, and apply.
+
+For active chat/session visibility today:
+
+```bash
+llm-wb sessions list
+```
+
+## Promote To Main
+
+Promotion is governed by:
+
+```txt
+.agentic/00.chat/workflows/chat-promote-to-main.md
+```
+
+The harness treats local merge readiness separately from remote push. A push
+always needs separate explicit approval.
+
+The current CLI shortcut verifies readiness and performs only the local merge:
+
+```bash
+llm-wb merge-main
+```
+
+## Report Or Close A Chat
+
+Reporting and closeout are governed by:
+
+```txt
+.agentic/00.chat/workflows/chat-reporting.md
+.agentic/00.chat/workflows/chat-cleanup.md
+```
+
+These workflows help summarize work, inspect active workspaces, and clean up
+empty chat branches when it is safe.
+
+`llm-wb list` intentionally lists installed workbench commands. Use
+`llm-wb sessions list` to list active chat sessions.

package/examples/minimal-repo/README.md

@@ -0,0 +1,13 @@
+# Minimal Target Repo
+
+This example represents the smallest kind of repo that can install the
+workbench:
+
+- it is a Git repo
+- it has no existing harness
+- it can receive `AGENTS.md`, `.agentic/`, `scripts/`, and `package.json`
+- its first chat startup creates its own `commitLogs/` inside a chat-owned
+  worktree
+
+The install smoke test creates a throwaway repo with this shape rather than
+modifying this example directory in place.

package/package.json

@@ -0,0 +1,93 @@
+{
+  "name": "llm-wb",
+  "version": "0.1.0-beta.0",
+  "description": "A portable chat harness for governed agentic development workflows.",
+  "license": "MIT",
+  "author": "Gordon Rose",
+  "homepage": "https://github.com/gordonrose/llm-workbench#readme",
+  "repository": {
+    "type": "git",
+    "url": "git+https://github.com/gordonrose/llm-workbench.git"
+  },
+  "bugs": {
+    "url": "https://github.com/gordonrose/llm-workbench/issues"
+  },
+  "keywords": [
+    "agentic",
+    "chat",
+    "cli",
+    "git",
+    "llm",
+    "workbench"
+  ],
+  "engines": {
+    "node": ">=18"
+  },
+  "publishConfig": {
+    "access": "public"
+  },
+  "bin": {
+    "llm-wb": "bin/llm-workbench.js"
+  },
+  "files": [
+    ".agentic/00.chat/",
+    ".agentic/shared/",
+    ".cursor/rules/llm-workbench.mdc",
+    ".github/copilot-instructions.md",
+    "AGENTS.md",
+    "CLAUDE.md",
+    "CONTRIBUTING.md",
+    "LICENSE",
+    "LLM_WORKBENCH.md",
+    "README.md",
+    "SECURITY.md",
+    "bin/",
+    "docs/00.chat/README.md",
+    "docs/00.chat/llm-workbench-acceptance-matrix.md",
+    "docs/00.chat/script-layout.md",
+    "docs/adapting-to-your-repo.md",
+    "docs/concepts.md",
+    "docs/install.md",
+    "docs/public-beta-contract.md",
+    "docs/workflows.md",
+    "examples/",
+    "scripts/00.chat/README.md",
+    "scripts/00.chat/bootstrap/",
+    "scripts/00.chat/closeout/",
+    "scripts/00.chat/command/",
+    "scripts/00.chat/git/",
+    "scripts/00.chat/local-merge/",
+    "scripts/00.chat/main-refresh/",
+    "scripts/00.chat/metrics/",
+    "scripts/00.chat/migration/",
+    "scripts/00.chat/recovery/",
+    "scripts/00.chat/reporting/",
+    "scripts/00.chat/session-log/",
+    "scripts/00.chat/startup/",
+    "scripts/00.chat/transcript/",
+    "scripts/00.chat/worktree/",
+    "scripts/01.harness/artifact-metadata/check-headers/script.sh",
+    "scripts/01.harness/artifact-metadata/check-headers/smoke-test.sh",
+    "scripts/01.harness/check-deterministic-process-drift.sh",
+    "scripts/01.harness/check-governed-script-command-drift.sh",
+    "scripts/01.harness/run-governed-script.sh",
+    "scripts/install.sh",
+    "scripts/uninstall.sh",
+    "tests/smoke-test-install.sh"
+  ],
+  "scripts": {
+    "chat": "bash scripts/00.chat/command/dispatcher/script.sh",
+    "chat:list": "bash scripts/00.chat/command/dispatcher/script.sh list",
+    "chat:new": "bash scripts/00.chat/command/dispatcher/script.sh new",
+    "chat:open-window": "bash scripts/00.chat/command/dispatcher/script.sh open-window",
+    "chat:close": "bash scripts/00.chat/command/dispatcher/script.sh close",
+    "chat:audit-bootstrap": "bash scripts/00.chat/bootstrap/audit-chat-bootstrap-file-set/script.sh",
+    "chat:audit-layer-migration": "bash scripts/00.chat/migration/audit-chat-layer-migration/script.sh",
+    "chat:cleanup-empty-branches": "bash scripts/00.chat/git/cleanup-empty-chat-branches/script.sh",
+    "chat:commit-log-summary": "bash scripts/00.chat/reporting/generate-commit-log-summary/script.sh",
+    "chat:record-main-refresh-conflict": "bash scripts/00.chat/session-log/record-main-refresh-conflict/script.sh",
+    "chat:report-workspaces": "bash scripts/00.chat/reporting/report-chat-workspaces/script.sh",
+    "chat:smoke-package-scripts": "bash scripts/00.chat/command/package-scripts/smoke-test.sh",
+    "test:install": "bash tests/smoke-test-install.sh"
+  }
+}

package/scripts/00.chat/README.md

@@ -0,0 +1,46 @@
+<!-- agentic-artifact:
+  schema: agentic-artifact/v2
+  id: chat.script.readme
+  version: 1
+  status: active
+  layer: 00.chat
+  domain: governance
+  disciplines:
+  - agentic
+  kind: script-domain-readme
+  purpose: Explain the canonical script layout for the chat lifecycle layer.
+  portability:
+    class: required
+    targets:
+    - llm-workbench
+  used_by:
+  - id: chat.readme
+    path: .agentic/00.chat/README.md
+  - id: harness.architecture.adr.0017-organize-scripts-by-owner-domain-and-capability
+-->
+# 00.chat Scripts
+
+This tree contains the canonical executable surface for the chat lifecycle
+layer. It is the script side of `.agentic/00.chat/`: workflows describe the
+governed process, and scripts provide deterministic checks, commands, reports,
+startup, recovery, and bookkeeping.
+
+Each capability lives in a folder:
+
+```txt
+scripts/00.chat/<domain>/<capability>/
+```
+
+The common shape is:
+
+- `script.sh` or `script.js`: the canonical implementation
+- `smoke-test.sh`: focused behavior proof, when the capability needs one
+- `README.md`: human onboarding for what the capability owns
+
+Public users should usually enter through `package.json` `chat:*` scripts.
+Workflow and gate authors should point to the canonical paths in this tree.
+
+Retired compatibility paths under `scripts/shared/chat/` and
+`scripts/shared/git/` should not be reintroduced. Historical documentation may
+mention them as previous locations, but live instructions should point here.
+

package/scripts/00.chat/bootstrap/README.md

@@ -0,0 +1,35 @@
+<!-- agentic-artifact:
+  schema: agentic-artifact/v2
+  id: chat.script.bootstrap.readme
+  version: 1
+  status: active
+  layer: 00.chat
+  domain: bootstrap
+  disciplines:
+  - agentic
+  kind: script-domain-readme
+  purpose: Explain bootstrap scripts for preparing a portable chat workbench repo.
+  portability:
+    class: required
+    targets:
+    - llm-workbench
+  used_by:
+  - id: chat.workflows.bootstrap-chat-workbench-repo
+    path: .agentic/00.chat/workflows/bootstrap-chat-workbench-repo.md
+  - id: chat.script.bootstrap.audit-chat-bootstrap-file-set.readme
+    path: scripts/00.chat/bootstrap/audit-chat-bootstrap-file-set/README.md
+-->
+# Bootstrap Scripts
+
+Bootstrap scripts help separate the portable chat workbench from this source
+repo. They answer a practical question: which files are required when creating
+or refreshing a standalone `llm-workbench` style repository?
+
+Bootstrap is not normal chat startup. It is source-to-upstream packaging work.
+It must avoid source-specific product files, session logs, transcripts, local
+paths, and deployment material.
+
+Use this domain when preparing the public workbench file set, validating that no
+required script is missing, or checking whether a script is merely a validation
+candidate.
+