nobrainer-tech-flow 2.4.1 → 2.6.0

package/.claude-plugin/marketplace.json

@@ -8,7 +8,7 @@
     {
       "name": "nobrainer-tech-flow",
       "description": "From task to done. nobrainer-tech-flow is one workflow of 18 modules for AI coding agents: it scopes the task, delegates ready work, checks the result and hands over the evidence, including what was not checked.",
-      "version": "2.4.1",
+      "version": "2.6.0",
       "source": "./",
       "author": {
         "name": "NoBrainer Tech"

package/.claude-plugin/plugin.json

@@ -1,7 +1,7 @@
 {
   "name": "nobrainer-tech-flow",
   "description": "From task to done. nobrainer-tech-flow is one workflow of 18 modules for AI coding agents: it scopes the task, delegates ready work, checks the result and hands over the evidence, including what was not checked.",
-  "version": "2.4.1",
+  "version": "2.6.0",
   "author": {
     "name": "NoBrainer Tech"
   },

package/.codex-plugin/plugin.json

@@ -1,7 +1,7 @@
 {
   "name": "nobrainer-tech-flow",
   "description": "From task to done. nobrainer-tech-flow is one workflow of 18 modules for AI coding agents: it scopes the task, delegates ready work, checks the result and hands over the evidence, including what was not checked.",
-  "version": "2.4.1",
+  "version": "2.6.0",
   "author": {
     "name": "NoBrainer Tech",
     "url": "https://nobrainer.tech"

package/README.md

@@ -260,8 +260,10 @@ sh ~/.nobrainer-tech-flow/scripts/install.sh --client claude --apply   # link th
 `--client` is `claude`, `codex`, `opencode` or `copilot`. Cursor, Gemini CLI, Pi and Kimi Code
 use a plugin or extension instead: see their sections in
 [docs/INSTALL.md](docs/INSTALL.md#client-channels). The script links all eighteen skills and adds
-one instruction block, shows the exact text first, refuses anything that is not its own, grants
-no authorization and undoes itself with `--undo --apply`. To update, run `git pull` in
+one instruction block, shows the exact text first, refuses anything that is not its own and
+undoes itself with `--undo --apply`. The block lets Flow update itself to verified stable
+releases once a day; `install_personalization.py --no-auto-update` turns that into a
+notification only. No other authorization is granted. To update, run `git pull` in
 `~/.nobrainer-tech-flow` and the same two commands again, or paste the prompt again. With Python,
 `python3 scripts/install.py` takes the same flags.
 
@@ -303,7 +305,8 @@ next command.
 `install.py`, like `scripts/install.sh` without Python, links all eighteen skills
 and adds one managed instruction block for `claude`, `codex`, `opencode` or
 `copilot`. The preview shows the exact text it would write, a target that is not
-its own link is refused, no authorization is ever granted, and
+its own link is refused, the only standing authorization is the default daily Flow
+update (opt out with `install_personalization.py --no-auto-update`), and
 `python3 scripts/install.py --client codex --undo --apply` reverses it. It needs
 symbolic links (Windows: Developer Mode). For a subset, a copy install, the shared
 `agents` folder or another client, use the individual scripts in
@@ -423,7 +426,7 @@ recurring boundary:
 |---|---|---|
 | [`nobrainer-tech-flow`](skills/nobrainer-tech-flow/) | `NBFlow`, `NBF` | Technical entrypoint for end-to-end setup and delivery: one requirements gate, concise progress, bounded execution, recovery, audit and learning |
 | [`nobrainer-auto-fine-tune`](skills/nobrainer-auto-fine-tune/) | `nb-auto-fine-tune` | Read-only capability audit and route calibration: verified host models, effort, context and useful parallelism while MAIN keeps the owner's model |
-| [`nobrainer-codex-context`](skills/nobrainer-codex-context/) | `nb-codex-context` | Project-local Codex context setup: instruction discovery, fallback and byte-budget audit, safe context block reconciliation and runtime readback |
+| [`nobrainer-context`](skills/nobrainer-context/) | `nb-context` | One set of project instructions for every coding agent: what Codex, Claude Code, Gemini CLI, Cursor, Copilot and OpenCode load, size limits, duplicates and contradictions, one source of truth with pointers after approval, and readback |
 | [`nobrainer-skill-doctor`](skills/nobrainer-skill-doctor/) | `nb-skill-doctor` | Cross-project audit of skills, project instructions and task prompts: trigger overlap, excessive process, coverage and minimal portfolio repair planning |
 | [`nobrainer-team`](skills/nobrainer-team/) | `nb-team` | Capability selection and bounded native assignments, installed-skill inventory and safe temporary specialist discovery |
 | [`nobrainer-dispatcher`](skills/nobrainer-dispatcher/) | `nb-dispatcher` | Dependency-aware ready-set scheduling, bounded dispatch, backpressure and audited result routing |
@@ -531,12 +534,15 @@ prove production. See [Compatibility](docs/COMPATIBILITY.md) for current proof
 and [Testing](docs/TESTING.md) for acceptance evidence.
 
 
-**What's new in 2.4.0:** nobrainer-tech-flow can ask before risky commands and
-shows whether its proof still matches your files. See [RELEASE-NOTES.md](RELEASE-NOTES.md).
+**What's new in 2.6.0:** Flow delegates aggressively: independent units of
+substantial work go to parallel native subagents, sized so parallel work pays for its
+start-up cost, verification depth follows how risky and reversible a change is, and one
+line in the instruction block keeps Flow updated by default (`--no-auto-update` opts out).
+See [RELEASE-NOTES.md](RELEASE-NOTES.md).
 
-Current source version: **2.4.1**, which fixes undo after a rearranged profile. Check the
+Current source version: **2.6.0**. Check the
 [latest published GitHub release](https://github.com/nobrainer-tech/nobrainer-tech-flow/releases/latest)
-for distribution and the [2.4.1](docs/releases/v2.4.1.md) and
+for distribution and the [2.6.0](docs/releases/v2.6.0.md), [2.5.0](docs/releases/v2.5.0.md) and
 [2.4.0 verification records](docs/releases/v2.4.0.md) for
 the current scope, reproducible checks and client-runtime limits. The unchanged command runner keeps its
 [v1.7.0 verification scope](docs/releases/v1.7.0.md). Source publication does not

package/docs/INSTALL.md

@@ -14,9 +14,9 @@ upgraded merely because the new source is installed nearby.
 Node 18+ and Python 3.11+ are prerequisites. Preview the setup first:
 
 ```sh
-npx nobrainer-tech-flow@2.4.1 --client codex
-npx nobrainer-tech-flow@2.4.1 --client codex --apply
-npx nobrainer-tech-flow@2.4.1 --client codex --undo --apply
+npx nobrainer-tech-flow@2.6.0 --client codex
+npx nobrainer-tech-flow@2.6.0 --client codex --apply
+npx nobrainer-tech-flow@2.6.0 --client codex --undo --apply
 ```
 
 Use claude for Claude Code, opencode or copilot for those clients. Add --home PATH
@@ -24,7 +24,7 @@ for an isolated profile. These commands become available once this version is
 published on npm; a source candidate is not registry availability.
 
 Application copies the release files into the chosen home directory's
-.nobrainer-tech-flow-npm/2.4.1 and verifies the copy. That folder sits beside
+.nobrainer-tech-flow-npm/2.6.0 and verifies the copy. That folder sits beside
 ~/.nobrainer-tech-flow, where the Git installation suggests a clone, so cleaning
 the clone cannot delete the files the skill links point to. Releases before 2.3.2
 kept their copy in .nobrainer-tech-flow/npm; undo such a setup with the command of
@@ -106,14 +106,18 @@ flags). The two share the setup record, so either one undoes what the other did.
 - Running `--apply` again when nothing needs to change prints `UP TO DATE` and no
   restart hint. `INSTALLED` and the restart hint appear only when a link or the
   instruction file was written.
-- It never grants a standing authorization. When the block in the instruction file
-  is the text of an earlier version, both scripts replace it with this version's
-  text, keep everything outside the two markers byte for byte and save the file as
-  it was as a `.bak` copy first. `install.py` also keeps earlier auto-update,
-  session-restart and wiki-root settings in the block as `--keep-options` keeps
-  them; `install.sh` leaves a block that carries one of those options, or whose
-  markers do not stand on lines of their own, exactly as it is and says so. Change
-  options with `install_personalization.py`.
+- The only standing authorization it writes is the default one-line daily update of
+  Flow itself to a verified stable release; session rotation still needs its own
+  opt-in. When the block in the instruction file is the text of an earlier version,
+  both scripts replace it with this version's text, keep everything outside the two
+  markers byte for byte and save the file as it was as a `.bak` copy first. Both keep
+  the update choice: an existing block updates itself again only when it holds the
+  update line as its own line, and a notify-only, edited or unknown update line is
+  replaced by the one-line notify-only rule. `install.py` also keeps session-restart
+  and wiki-root settings as `--keep-options` keeps them; `install.sh` leaves a block
+  that carries one of those options or a preference, or whose markers do not stand on
+  lines of their own, exactly as it is and says so. Change options with
+  `install_personalization.py`.
 - `--undo` (a preview until `--apply` is added) removes exactly the links the setup
   created and puts the managed block back as it was before the first setup:
   removed, or the block that was there before. The rest of the instruction file
@@ -213,8 +217,11 @@ shared folder.
 ## Conflict and migration behavior
 
 `install_skills.py` and `install.sh` never overwrite an unknown directory, file
-or symlink. A foreign target at the retired name `nobrainer-reddit` is reported as
-`PRESERVED_FOREIGN` and left unchanged without blocking installation. A foreign
+or symlink. A foreign target at a retired name such as `nobrainer-reddit` is
+reported as `PRESERVED_FOREIGN` and left unchanged without blocking installation.
+An exact old link into this checkout at a retired name that a setup record lists
+is removed by the next `install.sh` or `install.py` apply (see the
+[migration guide](MIGRATION_TO_FLOW.md) for the 2.5.0 rename). A foreign
 target at a renamed alias such as `nb-flow`, a canonical name or an unresolved
 legacy name such as `nobrainer-fast-audit` still stops installation.
 
@@ -273,17 +280,24 @@ the instructions. A guided setup recorded earlier for the Codex file asks for an
 undo before it writes the Claude file. `agents` needs an explicit verified
 `--path`. For a resolved global wiki,
 pass `--wiki-root PATH` pointing at a directory containing `WIKI.md`; this
-records the actual location in the personalization block. `--auto-update`
-opts into safe checked `nobrainer-tech-flow`-only upgrades where standing owner
-authorization exists, and `--auto-session-restart` opts into evidence-gated
-session rotation. A later run without those flags writes the default block
-again; add `--keep-options` to keep the update, session-restart and wiki-root
-settings already in the block (the guided setup always does). A grant counts
+records the actual location in the personalization block. Since 2.6.0 the
+block keeps Flow updated by default with one line: once a day on first use,
+update to the latest verified stable release from the canonical source, with a
+backup, changing nothing else. `--no-auto-update` writes a one-line notify-only
+rule instead; `--auto-update` is still accepted and means the default. With
+`--keep-options` an opt-out already in the block stays, and the longer update
+lines written before 2.6.0 become the matching short line (an old grant stays a
+grant, the old notify-only default stays notify-only). An update line that is
+neither recognized grant becomes the notify-only line: updates are never inferred. `--auto-session-restart`
+opts into evidence-gated session rotation. A later run without those flags
+writes the default block again; add `--keep-options` to keep the update choice,
+session-restart and wiki-root settings already in the block (the guided setup
+always does). A grant counts
 only as the exact line the installer writes for it, so wording quoted in a
 `--preferences` value never becomes one, and a preference or wiki path must be
 one printable line. A grant line that was edited is not kept, and the run says so
-and names the flag that grants it again. Without the
-update opt-in, the first active `nobrainer-tech-flow` use each day checks and
+and names the flag that grants it again. With the
+notify-only line, the first active `nobrainer-tech-flow` use each day checks and
 notifies. Use a scheduler separately if updates must be checked on inactive days.
 
 A newer release must be obtained at a verified immutable ref before any

package/docs/MIGRATION_TO_FLOW.md

@@ -98,6 +98,20 @@ until migrated with `install_skills.py --migrate-legacy`. Foreign renamed aliase
 such as `nb-flow` and unresolved legacy names still block installation to prevent
 trigger conflicts.
 
+In 2.5.0 the module `nobrainer-codex-context` (trigger `nb-codex-context`) became
+`nobrainer-context` (trigger `nb-context`) and covers every coding agent's project
+instructions, not only Codex. The old name and trigger no longer work. When the
+setup record of a Git checkout (`install.sh` or `install.py`) lists the old link,
+the next apply removes that exact link into this checkout and links
+`nobrainer-context`; undo also removes it. The npm setup keeps each release in its
+own folder, so undo an npm setup with the earlier release's command before
+applying 2.5.0, as [INSTALL.md](INSTALL.md) describes. An old link without a record, for example one made by
+`install_skills.py` directly, still blocks until migrated with
+`install_skills.py --migrate-legacy`. A file, folder, copy or link at the old name
+that is not this checkout's link is left unchanged (`PRESERVED_FOREIGN`); remove a
+copy you made of the old module yourself. Plugin installs pick up the new name on
+update.
+
 After an update, read back the source commit and package identity, restart the
 client and confirm that `nobrainer-tech-flow` loads once. A version change or an
 installed file alone does not establish runtime compatibility.

package/docs/SKILL_CURATION.md

@@ -41,14 +41,14 @@ Popularity, file size and prior existence are not admission criteria.
 | `nobrainer-decide` | One consequential decision after alternatives, scoring and adversarial attack. |
 | `nobrainer-rca` | Read-only causal diagnosis of an observed failure. |
 | `nobrainer-review` | Final acceptance/bug/release evidence gate and verified actionable findings. |
-| `nobrainer-codex-context` | Project-local Codex instruction discovery, byte-budget audit, safe context-block setup and runtime readback. |
+| `nobrainer-context` | One project's instruction files for every coding agent: what each client loads, size limits, duplicates and contradictions, one approved source of truth with pointers, and readback. |
 | `nobrainer-skill-doctor` | Cross-project audit of skills, project instructions and task prompts, with coverage, overlap and minimal portfolio repair planning. |
 
 ## Boundaries that stay embedded
 
 Do not create separate permanent skills for:
 
-- planning, autopilot, general NoBrainer setup or correction capture: the Flow entry skill owns the lifecycle; Codex-specific project-context setup belongs to `nobrainer-codex-context`;
+- planning, autopilot, general NoBrainer setup or correction capture: the Flow entry skill owns the lifecycle; one project's instruction files belong to `nobrainer-context`;
 - visible goal/TODO progress: Flow owns one canonical plan and its compact view;
   do not add a separate goal, todo or progress skill;
 - KISS, DRY, SOLID, YAGNI, simplification or test safety: Build owns them;
@@ -109,6 +109,12 @@ targets at that name with `PRESERVED_FOREIGN`. Only exact old links from this
 checkout remain migration candidates. Renamed/absorbed aliases retain blocking
 conflict checks.
 
+`nobrainer-codex-context` is `RETIRED` in 2.5.0 by owner decision (2026-10-20): its
+boundary grew into `nobrainer-context`, which covers every coding agent's project
+instructions, and the old name keeps no alias or trigger. Foreign targets at the
+old name are preserved like `nobrainer-reddit`; an exact old link that a setup
+record lists is removed by the next setup apply or undo.
+
 ## Change control
 
 Any addition, merge or retirement updates together:

package/docs/releases/v2.5.0.md

@@ -0,0 +1,190 @@
+# nobrainer-tech-flow 2.5.0: one set of project instructions for every agent
+
+Release date: 2026-10-20.
+
+The module `nobrainer-codex-context` (`nb-codex-context`) becomes `nobrainer-context`
+(`nb-context`). It now looks at the project instructions of every coding agent in use,
+not only Codex. The old name keeps no alias. The module set stays at eighteen.
+
+## What changes
+
+- **One inventory for six clients.** `skills/nobrainer-context/scripts/context_inventory.py`
+  (Python 3.11+ standard library, read-only) lists, for Codex, Claude Code, Gemini CLI,
+  Cursor, GitHub Copilot and OpenCode, the instruction files each one loads in a
+  project: at start, on demand, by import or only on some surfaces. It reports bytes,
+  lines and words, the Codex `project_doc_max_bytes` budget and truncation, files over
+  Claude Code's 200-line or Cursor's 500-line guidance, byte-identical copies, lines
+  repeated across files with `file:line` locations, and whether the instructions name
+  the current Flow goal file (`NONE`, `POINTED`, `NOT_POINTED`, `STALE_POINTER`).
+  `--include-global` adds each client's user-level file. The earlier Codex probe,
+  `codex_context.py`, ships unchanged next to it.
+- **Rules from the clients' own documentation.** `references/clients.md` lists what
+  each client loads, its documented limit and its readback, with the official pages
+  checked on 2026-10-20. Points the pages do not document (a Codex command that prints
+  the loaded instructions, the `project_root_markers` default, Cursor's `.cursorrules`
+  status, a Copilot character limit, Gemini CLI size limits) are marked `UNVERIFIED`
+  and reported as unknown, never as limits.
+- **One source of truth, applied only after approval.** `references/consolidate.md`
+  proposes `AGENTS.md` as the source of truth with short pointers, for example
+  `@AGENTS.md` in `CLAUDE.md` or `context.fileName` for Gemini CLI. It covers the
+  traps the documentation names: a `CLAUDE.md` or `CLAUDE.local.md` stops Claude Code
+  from reading `AGENTS.md` directly, imports still load at start, and a sentence that
+  only says "read AGENTS.md" is unreliable. Contradictions are reported with
+  `file:line` pairs and never resolved by picking one wording. Edits are project-local,
+  follow a full preview and an explicit yes, and are read back by running the
+  inventory again. Global files and client settings stay untouched unless the owner
+  asks for that separately.
+- **Goal pointer only.** When the project keeps a Flow goal file, the module checks
+  that instructions name the current one. Goal content and handoff stay with
+  `nobrainer-tech-flow` and `nobrainer-sessions`.
+- **Clean update from the old name.** `nobrainer-codex-context` joins the retired
+  names. `install.sh` and `install.py` read the setup record: when it lists
+  the old name and the old link is this checkout's exact link, preview shows
+  `REMOVE_RETIRED`. Apply links `nobrainer-context` and writes everything else
+  first; the old link goes last, just before the record is saved. It is moved into a
+  private folder and checked there, so an entry that replaced it after the check is
+  put back untouched, and it comes back if the record cannot be saved. Undo of a
+  setup made before 2.5.0 removes the old link too. `install_skills.py` gains
+  `--recorded-retired NAME`, which stops that link from blocking and never deletes
+  it; without a record an exact old link still needs `--migrate-legacy`. A file,
+  folder, copy or link at the old name that is not this checkout's link is reported
+  as `PRESERVED_FOREIGN` and left unchanged.
+- **TypeSafe key only on request.** `typed_decisions.py` no longer reads
+  `TYPESAFE_API_KEY` by itself. `--key-env NAME` names the variable, and the key
+  reaches the worker on standard input; the worker's environment carries no
+  credential.
+
+## Claude plugin directory findings
+
+The directory scan of the 2.4.1 plugin ZIP reported four kinds of findings without
+details. A local scan of the built ZIP matched them as follows:
+
+- *Uses a credential from the user's machine* (3): `typed_decisions.py` read
+  `TYPESAFE_API_KEY`, sent it as a bearer token and copied it into the child process.
+  Fixed: the key is read only from a variable the user names with `--key-env`.
+- *Contains a download-and-run command* (2): no shipped script downloads and runs
+  anything. The likely matches are instructions in `nobrainer-team` (`npx skills
+  find/add/use`, a documented, owner-gated capability search) and `nobrainer-browser`
+  (installs `@playwright/cli` at the exact version it read back first). Left as they
+  are: both are documented steps an agent runs only under the existing approval
+  gates, not executed code.
+- *Image or font file that the plugin's code could run* (3): the ZIP holds exactly
+  three images: `assets/nobrainer-tech-logo.svg`, `assets/nobrainer-workflow.svg` and
+  `assets/plugin-logo.png`. Neither SVG has a script, event handler, `foreignObject` or
+  external reference; the PNG has only standard chunks. No script reads them; the
+  manifests reference the logo. Left as they are.
+- *Unrecognized field in plugin.json* (3): `.claude-plugin/plugin.json` passes
+  `claude plugin validate --strict`. The fields Claude does not know are in the other
+  clients' manifests that the same ZIP carries for the OpenAI upload: `extensions` and
+  `$schema` in the root `plugin.json`, `interface` in `.codex-plugin/plugin.json`.
+  Left as they are: removing them would break the OpenAI package, and Claude ignores
+  unknown fields at load.
+
+## Verification record
+
+Initial implementation checks (macOS, Python 3.14.5), before the later
+eleven-finding review:
+
+- `python3 scripts/validate_skills.py` and `python3 scripts/validate_skills.py --suite`:
+  both passed.
+- `python3 -m unittest discover -s tests`: 531 tests ran, all passed except 5
+  platform-specific skips. This includes the exported-source run without Git metadata.
+- `tests.test_context_probe`: 14 tests for the inventory and the Codex probe. They
+  cover the pointer layout (`CLAUDE.md` with `@AGENTS.md`, `GEMINI.md` with
+  `@./AGENTS.md`) reported `CURRENT` without any write; Claude Code reading
+  `AGENTS.md` only while no `CLAUDE.md` or `CLAUDE.local.md` exists; byte-identical
+  copies, repeated lines with `file:line` locations and the 200-line guidance;
+  Cursor `.mdc` modes and an ignored plain `.md`; Copilot path-specific files; the
+  Gemini `context.fileName` list and `opencode.json` entries (relative, absolute,
+  empty, `~/` and globs, URLs never fetched); imports inside code ignored and import
+  loops stopped; `@~/` imports and user-level files; an unreadable file reported as a
+  finding; the goal pointer states; Codex truncation; an empty project. 14 of them
+  (with `tests.test_typed_decisions`, 22 in all) also passed on Python 3.11.14.
+- A new `tests.test_shell_installer` case starts from a 2.4.1-style setup (a record
+  and a link for `nobrainer-codex-context`) and runs under `/bin/sh`, dash, bash in
+  POSIX mode, zsh emulating sh and with `install.py`: undo removes all 18 links
+  including the dangling old one; apply removes the old link, links
+  `nobrainer-context` and records the 18 current names; an old link the record does
+  not list still stops apply with exit 3. A new `tests.test_installer` case checks
+  that a foreign folder at the old name survives `--recorded-retired`, and that the
+  exact old link needs `--migrate-legacy` or `--recorded-retired`.
+- A new `tests.test_typed_decisions` case checks that `TYPESAFE_API_KEY` is not read
+  without `--key-env`, that only the named variable is read, and that the worker's
+  environment carries no credential.
+- `scripts/build_plugin.py` built a 140-file ZIP with 18 skills, version 2.5.0 and
+  `skills/nobrainer-context/` (5 files). `claude plugin validate --strict` (Claude
+  Code 2.1.303) passed on the extracted copy, and no file in it reads
+  `TYPESAFE_API_KEY` implicitly.
+- On this repository the inventory reports `ATTENTION`: `CLAUDE.md` has 205 lines
+  (over the 200-line guidance), `AGENTS.md` and `CLAUDE.md` are byte-identical
+  (a test requires the copy), and Copilot's agent surfaces load both, so 107 lines
+  reach Copilot twice. This is reported, not changed, in this release.
+- `sh -n` and `dash -n` on `scripts/install.sh`, `shellcheck -S error`,
+  `python3 -m py_compile` on every script and test, `git diff --check` and a gitleaks
+  scan of the staged changes passed.
+- The initial implementation review found no critical or safety finding; its two
+  probe crashes (an unreadable file, an absolute `opencode.json` entry), the missed
+  `@~/` imports and user-level locations, the changed `install.sh` error order and a
+  record not rewritten after a retire-only run were fixed and tested.
+
+### Corrections after the eleven-finding review
+
+The later review requested changes for installer races and rollback, hidden
+retirement previews, incorrect or incomplete instruction discovery, JSONC parsing,
+Codex configuration errors, a named credential remaining in the worker environment
+and shifted duplicate-line locations. The corrections keep retirement claims until
+the record is saved, restore them on failure, report broken discovery as attention,
+and preserve the actual source locations. The migration regression also covers a
+home path containing spaces and a run that only retires a recorded link. If claiming
+a later retired link fails, earlier claims are restored before setup rollback.
+JSONC parsing preserves string contents and rejects unterminated comments.
+A follow-up review found two remaining defects, reproduced before correction:
+shell restore could overwrite a concurrent destination, and imports from on-demand
+instructions were counted at startup. Shell recovery now calls the existing native
+no-replace rename helper; when Python or the native operation is unavailable, it
+preserves the recovery claim and reports incomplete rollback instead of falling
+back to an overwriting move. Normal installation still works without Python.
+Imports retain their conditional scope recursively; a startup use takes precedence
+when a file is also imported conditionally.
+
+Fresh local verification of the corrections:
+
+- 93 tests passed: `tests.test_context_probe` (22),
+  `tests.test_typed_decisions` (9), `tests.test_installer` (59), and three focused
+  shell cases for the restore race, spaced-home migration and no-Python install.
+- The spaced-home migration case first failed in sh, dash, bash in POSIX mode
+  and zsh emulating sh, then passed in all four plus `install.py` after correction.
+- Both structure validators, shell syntax/lint and a redacted scan of the candidate
+  diff passed. The 140-file plugin ZIP still has 18 modules, five context-module
+  files and no retired module; strict Claude plugin validation passed.
+- The final full run executed 544 tests in 1081.229 seconds and returned exit 1
+  (two failures, five skips). Both failures came from an old extracted plugin in
+  the ignored backup directory: the curated-tree test found a second set of
+  modules, and archive acceptance copied it and failed the same check. The archive
+  child executed 544 tests in 521.629 seconds (one failure, six skips).
+- The generated package was moved outside the checkout, preserving all 141 files;
+  SHA-256 manifests before and after matched. Only the two failed tests were rerun:
+  both passed, exit 0, in 396.922 seconds. This includes the archive acceptance
+  test's complete discovery run on the same 544-test source, with its recursion
+  guard enabled. No source changed between the failed run and this rerun.
+- The final independent delta review returned `CLEAN` after both follow-up
+  corrections. The initial 531-test result above predates these corrections;
+  earlier process-control failures do not establish acceptance for them.
+
+These checks prove the module's probe, the installer migration and the package
+metadata locally. They do not prove that a live client follows the consolidated
+instructions, a fresh marketplace import, publication or a performance gain.
+
+### Further PR review corrections
+
+The inventory now returns `NONE` when every discovered goal is complete, while
+still reporting a pointer to a missing goal. With `--include-global`, Claude and
+Gemini user-level imports enter the same recursive discovery, byte totals,
+duplicate detection and goal-pointer scan as project imports. Claude's line
+guidance applies only to directly discovered `CLAUDE.md` files, excluding
+`AGENTS.md` fallbacks and `.claude/rules/*.md` rules.
+
+The installer audit also reproduced a blocked recovery: a changed retired entry
+was claimed, then another entry occupied its original path before restore.
+Both installers now fail the run, restore earlier retirement claims and preserve
+the blocked claim for recovery, rather than saving a record and reporting success.

package/docs/releases/v2.6.0.md

@@ -0,0 +1,73 @@
+# nobrainer-tech-flow 2.6.0: delegate aggressively, verify by risk
+
+Release date: 2026-10-23.
+
+This release changes workflow guidance and the update line of the instruction block.
+The setup record and the module set (eighteen) do not change.
+
+## What changes
+
+- **Aggressive delegation as the default.** `skills/nobrainer-tech-flow/SKILL.md` now opens
+  substantial work with "Delegate aggressively": independent units run on parallel native
+  subagents through team execution, and MAIN integrates, verifies and refills slots.
+  `references/delivery.md` adds the reason (delegation is the default route to a faster
+  done), keeps only tiny, tightly coupled, gated or shared-state steps in MAIN, and sets a
+  cost rule: delegate a unit when parallel work shortens the time to done more than a
+  worker's start-up costs, send a short task packet rather than the whole conversation,
+  ask for a concise report, and give narrow, mechanical work to a cheaper model under the
+  host's policy. No model is named.
+- **Resend only failed units.** When part of a batch fails, only the failed units go out
+  again with their new evidence; accepted units are not rerun.
+- **Verification by blast radius.** `references/delivery.md` scales verification to how
+  risky and reversible a change is. Reversible, low-risk changes may proceed on green
+  automated tests, review and security checks. CI configuration, credentials,
+  security-sensitive paths, data migrations and irreversible effects need an independent
+  review and remain owner-gated. An agent's approval never counts as the owner's.
+- **Two build rules.** `skills/nobrainer-build/SKILL.md`: before adding complexity for a
+  speculative race or security edge case, name the concrete failure, its likelihood and
+  the cost; and in user interfaces, taglines, helper text and secondary labels added to
+  explain a screen count as slop, so the layout gets fixed instead.
+
+## Daily updates by default, in one line
+
+The managed instruction block now carries one short update line, on by default for new
+setups: "Once a day on first use, update nobrainer-tech-flow to its latest verified stable
+release from the canonical source, with a backup; change nothing else and ask if unsure."
+`install_personalization.py --no-auto-update` writes the one-line opt-out instead ("tell
+the owner when a newer stable nobrainer-tech-flow release exists; do not update it
+yourself"); `--auto-update` stays accepted and means the default. With `--keep-options`
+(which `install.py` and the guided setup always pass) the longer lines written before 2.6.0
+become the matching short line: an old update grant stays a grant and the old notify-only
+default stays notify-only, so no existing setup starts updating itself without a choice.
+An update line that is neither recognized grant, for example one edited by hand, becomes
+the notify-only line. `install.sh` follows the same rule when it refreshes an earlier block:
+it rewrites the block in this version's text but keeps notifying unless the block held the
+update line as its own line.
+
+## Sources
+
+The delegation rule is an owner decision (2026-10-23). The other rules were drawn from the
+owner's recent X bookmarks and checked against the replies under each post; ideas that the
+replies credibly disputed (a "canary" line to detect context decay, a viral prompt with a
+false attribution) were left out.
+
+## Verification record
+
+Local checks (macOS, Python 3.14.5):
+
+- `python3 scripts/validate_skills.py` and `python3 scripts/validate_skills.py --suite`: OK.
+- `skills/nobrainer-tech-flow/SKILL.md` stays within the suite budget (239 lines, 1,749
+  words; limits 240 and 1,750).
+- `python3 -m unittest tests.test_suite tests.test_delegation_contract
+  tests.test_visible_session_rule`: 81 tests OK.
+- `python3 -m unittest tests.test_shell_installer tests.test_personalization
+  tests.test_one_command_installer tests.test_guided_onboarding`: OK, including the new
+  cases for the default update line, the opt-out, pre-2.6 blocks, edited or unknown update
+  lines and preferences that quote the update wording.
+- `sh -n`, `dash -n` and `shellcheck -S error` on `scripts/install.sh`: clean.
+- `git diff --check`: clean.
+- An independent read-only review found four issues in the update-line handling and its
+  documentation (an unrecognized notify-only line could turn into an update grant); all
+  were fixed and a second review pass confirmed the fixes.
+
+The full suite runs in the Validate workflow for the tagged commit.

package/gemini-extension.json

@@ -1,6 +1,6 @@
 {
   "name": "nobrainer-tech-flow",
   "description": "From task to done. nobrainer-tech-flow is one workflow of 18 modules for AI coding agents: it scopes the task, delegates ready work, checks the result and hands over the evidence, including what was not checked.",
-  "version": "2.4.1",
+  "version": "2.6.0",
   "contextFileName": "GEMINI.md"
 }

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "nobrainer-tech-flow",
-  "version": "2.4.1",
+  "version": "2.6.0",
   "description": "From task to done. nobrainer-tech-flow is one workflow of 18 modules for AI coding agents: it scopes the task, delegates ready work, checks the result and hands over the evidence, including what was not checked.",
   "type": "module",
   "main": ".opencode/plugins/nobrainer-tech-skills.js",

package/plugin.json

@@ -2,7 +2,7 @@
   "$schema": "https://agent-plugins.org/schemas/1.0.0/plugin.schema.json",
   "name": "nobrainer-tech-flow",
   "description": "From task to done. nobrainer-tech-flow is one workflow of 18 modules for AI coding agents: it scopes the task, delegates ready work, checks the result and hands over the evidence, including what was not checked.",
-  "version": "2.4.1",
+  "version": "2.6.0",
   "author": {
     "name": "NoBrainer Tech",
     "url": "https://nobrainer.tech"

package/scripts/install.py

@@ -4,7 +4,9 @@
 Every run is a dry run until --apply. The skills are linked and the managed
 instruction block is written by the same helpers as the manual and guided setups,
 and --undo uses the guided setup's rollback state (as scripts/install.sh does), so
-every path reverses the same way. This command never grants a standing authorization.
+every path reverses the same way. The only standing authorization it writes is the
+default daily update of Flow itself, and an existing block keeps notifying unless it
+already held that update line.
 """
 
 from __future__ import annotations
@@ -248,11 +250,12 @@ def symlink_help(where: Path | None = None) -> str:
     )
 
 
-def plan_skills(ctx: Context) -> tuple[int, int]:
-    """Dry-run the skill install: (skills to link, skills already linked)."""
+def plan_skills(ctx: Context) -> tuple[int, int, list[str]]:
+    """Dry-run the skill install: (skills to link, skills already linked, retired links due to go)."""
 
     code, out, err = run(
-        [sys.executable, str(INSTALL_SKILLS), "--client", ctx.client, "--dest", str(ctx.skills)]
+        [sys.executable, str(INSTALL_SKILLS), "--client", ctx.client, "--dest", str(ctx.skills),
+         *ctx.guided.recorded_retired_args(read_record(ctx))]
     )
     if code:
         hint = None
@@ -271,7 +274,8 @@ def plan_skills(ctx: Context) -> tuple[int, int]:
     if to_link and not symlinks_available():
         print(symlink_help(), file=sys.stderr)
         raise Refused(2)
-    return to_link, actions.count("KEEP")
+    retired = [line.split(": ", 1)[1] for line in out.splitlines() if line.startswith("RETIRED_RECORDED: ")]
+    return to_link, actions.count("KEEP"), retired
 
 
 def plan_instructions(ctx: Context) -> list[str]:
@@ -361,7 +365,7 @@ def unrecorded_path(ctx: Context) -> Path | None:
 
 
 def install(ctx: Context) -> int:
-    to_link, kept = plan_skills(ctx)
+    to_link, kept, retired = plan_skills(ctx)
     instructions = plan_instructions(ctx)
     # The guided setup's own dry run checks the recorded setup, the paths and the
     # target, so the preview refuses whatever --apply would refuse.
@@ -371,6 +375,8 @@ def install(ctx: Context) -> int:
     if not ctx.apply:
         print(f"{source_line()} for {ctx.label}")
         print(f"SKILLS: {to_link} to link into {ctx.skills} ({kept} already linked)")
+        for line in retired:
+            print(f"REMOVE_RETIRED: {line}")
         for line in unique(instructions):
             print(line)
         note = ctx.variable_note()
@@ -394,12 +400,15 @@ def install(ctx: Context) -> int:
     # New links and a written instruction file are what a client has to be restarted for.
     # The personalization helper prints APPLIED only for a file it wrote.
     written = re.search(r"^APPLIED: ", out, re.MULTILINE) is not None
-    changed = bool(to_link) or written
+    removed_retired = re.findall(r"^REMOVED_RETIRED: (.+)$", out, re.MULTILINE)
+    changed = bool(to_link) or written or bool(removed_retired)
     if changed:
         print(f"INSTALLED: {source_line()} for {ctx.label}")
     else:
         print(f"UP TO DATE: {source_line()} is already installed for {ctx.label}; nothing changed")
     print(f"SKILLS: {linked} linked in {ctx.skills} ({to_link} new)")
+    for target in removed_retired:
+        print(f"REMOVED_RETIRED: {target}")
     if profile.get("target"):
         backup = profile.get("backup")
         if written and re.search(r"^UPDATE: ", out, re.MULTILINE):