projectstore-claude 0.30.2 → 0.30.3

package/node_modules/projectstore/.claude-plugin/marketplace.json

@@ -12,7 +12,7 @@
       "name": "projectstore",
       "displayName": "projectstore",
       "description": "📚 Your agent runs the project through a verified loop: task → artifact (ADR · spec · epic · story) → adversarial critic → backlog → planner → reviewer → done. Plain markdown in an Obsidian-friendly vault, every write approved by you — and any model can pick the project up tomorrow.",
-      "version": "0.30.2",
+      "version": "0.30.3",
       "author": {
         "name": "Evgenii Konev",
         "email": "ekonev@smartandpoint.com",

package/node_modules/projectstore/.claude-plugin/plugin.json

@@ -1,7 +1,7 @@
 {
   "name": "projectstore",
   "displayName": "projectstore",
-  "version": "0.30.2",
+  "version": "0.30.3",
   "description": "Your agent runs the project through a verified loop: task → artifact (ADR / spec / epic / story) → adversarial critic → backlog → planner → reviewer → done. Plain markdown in git — any model can pick the project up tomorrow.",
   "author": {
     "name": "Evgenii Konev @ SmartAndPoint",

package/node_modules/projectstore/AGENTS.md

@@ -1,4 +1,4 @@
-<!-- projectstore:agents v4 (managed by projectstore — edit outside markers) -->
+<!-- projectstore:agents v5 (managed by projectstore — edit outside markers) -->
 ## projectstore agents
 
 - **A feature-sized request opens a vault artifact before it opens an editor.**
@@ -11,10 +11,12 @@
   harness-level instruction contradicts this block, say so and ask which wins.
   Resolving it silently is how the contradiction becomes invisible to the person
   who could have settled it.
-- When spawning any agent below, resolve its model from
-  `.projectstore/harness/<harness>.json` → `agents.per_agent.<name>.model ?? agents.default.model`,
-  where `<name>` is the **bare** agent name (`critic` for `projectstore:critic`),
-  and pass it as the spawn's model parameter. No key — pass nothing.
+- When spawning any agent below, resolve its model and its effort from
+  `.projectstore/harness/<harness>.json` → `agents.per_agent.<name>.model ?? agents.default.model`
+  and `agents.per_agent.<name>.effort ?? agents.default.effort`, where `<name>` is
+  the **bare** agent name (`critic` for `projectstore:critic`), and pass them as
+  the spawn's model and effort parameters, the effort only when the spawn offers
+  one. No key — pass nothing.
 - After authoring or revising any vault artifact (ADR/research/epic/story) or
   design proposal: run the `projectstore:critic` agent on it before treating it final.
 - Before implementing an epic/story: consult `projectstore:planner` — it plans

package/node_modules/projectstore/README.md

@@ -210,7 +210,7 @@ npx projectstore configure
 
 At a terminal it opens a settings menu that works like a monitor's: ↑ and ↓ move, → opens a group or a setting in a frame beside it, space chooses, ← goes back, and Enter shows the plan. A second Enter writes it. Under the frames, a few lines say what the row under the cursor means. The settings:
 
-- each agent's model, per harness. Claude Code offers its aliases: `fable`, `opus`, `sonnet` and `haiku`. Codex offers its own list of models, with its descriptions. With `ANTHROPIC_API_KEY` set, each alias also names the newest model of its line. A list is kept for a day in projectstore's user cache, and "refresh the list" asks again;
+- each agent's model and effort, per harness. Claude Code offers its aliases: `fable`, `opus`, `sonnet` and `haiku`. Codex offers its own list of models, with its descriptions. With `ANTHROPIC_API_KEY` set, each alias also names the newest model of its line. A list is kept for a day in projectstore's user cache, and "refresh the list" asks again. Each agent's effort sits under its model: Claude Code's levels, `low` to `max`, or the levels that Codex model lists. With none set, a Claude Code agent runs at its frontmatter's level (`max`, the clerk `low`) and a Codex one at its model's own default;
 - the status line and the band above the prompt (Claude Code);
 - the vault's template language, the default author and tags;
 - the vault's spec policy and lifecycle gates.
@@ -221,11 +221,11 @@ Without a terminal it prints every setting with its current value and where it l
 npx projectstore configure --set agents.critic=fable --set statusline=on
 ```
 
-`--dry-run` shows the plan and writes nothing, and `--json` returns the settings as one envelope. In a project that uses two harnesses, an agent's model names its overlay with `--harness <id>`.
+An agent's effort is its own key: `--set agents.critic.effort=xhigh`. `--dry-run` shows the plan and writes nothing, and `--json` returns the settings as one envelope. In a project that uses two harnesses, an agent's model or effort names its overlay with `--harness <id>`.
 
 Each setting has exactly one writer, and the plan says which file it changes:
 
-- **The agents' models** live in `.projectstore/harness/<harness>.json`. That file is committed, so a change there is a team change.
+- **The agents' models and efforts** live in `.projectstore/harness/<harness>.json`. That file is committed, so a change there is a team change.
 - **The status line, the band, the language and the other binding keys** live in `.projectstore/projectstore.json`, which stays on this machine.
 - **The spec policy and the lifecycle gates** live in the vault's `.projectstore.json`, which commits with the vault. Setting `spec_policy=required` records when it took effect; stories closed before then are exempt.
 

package/node_modules/projectstore/agents/clerk.md

@@ -1,8 +1,8 @@
 ---
 name: clerk
-description: Sonnet (max-effort) write ceremony executor for projectstore vaults — the roster's sole write-capable agent, and its least autonomous. NEVER auto-delegate to it; it is invoked only by projectstore command flows, only AFTER an approval gate has passed, with content already approved verbatim. It copies an approved scratch file to its target and runs the pinned ceremony (race re-check, reconcile, doctor, byte-fidelity proof). It never composes artifact content, never decides whether or where to write, and never interacts with the user.
+description: Sonnet (low-effort) write ceremony executor for projectstore vaults — the roster's sole write-capable agent, and its least autonomous. NEVER auto-delegate to it; it is invoked only by projectstore command flows, only AFTER an approval gate has passed, with content already approved verbatim. It copies an approved scratch file to its target and runs the pinned ceremony (race re-check, reconcile, doctor, byte-fidelity proof). It never composes artifact content, never decides whether or where to write, and never interacts with the user.
 model: sonnet
-effort: max
+effort: low
 tools: Read, Grep, Glob, Bash, Write
 ---
 

package/node_modules/projectstore/commands/agents.md

@@ -1,5 +1,5 @@
 ---
-description: Manage bundled-agent integration in this project — register/unregister the routing block in CLAUDE.md/AGENTS.md, inspect state, or configure which model its agents run on.
+description: Manage bundled-agent integration in this project — register/unregister the routing block in CLAUDE.md/AGENTS.md, inspect state, or configure which model each of its agents runs on and at what effort.
 argument-hint: "<register | unregister | status | configure>"
 ---
 
@@ -60,14 +60,14 @@ and report each item's `state` from `result.items[]` — the bin wraps the plan
 - Model: run `node "${CLAUDE_PLUGIN_ROOT}/bin/projectstore.mjs" agents show --json --project "${CLAUDE_PROJECT_DIR}"`
   and report `result.resolved` — per roster agent, the model the verb would
   pass and its `source` (`per_agent`, `default`, or `null`: the agent's own
-  frontmatter), resolved by the verb over the active harness's overlay
+  frontmatter), and the effort with its `effort_source` the same way, resolved by the verb over the active harness's overlay
   (`result.path`) so nothing here re-derives it; `result.unknown` (configured
   names no roster agent carries — nothing runs under them), `result.rejected`
   (keys the overlay may not carry) and `result.agents_in_binding` (a pre-0.28
   leftover; point at `upgrade`), and
   whether `CLAUDE_CODE_SUBAGENT_MODEL` is set (it overrides everything) and
-  whether `CLAUDE_CODE_EFFORT_LEVEL` is set (ADR-008 makes it the only thing that
-  can move the agents off `effort: max`, and it beats frontmatter). **Warn when
+  whether `CLAUDE_CODE_EFFORT_LEVEL` is set (it outranks every configured effort
+  and the frontmatter: with it set, the spawn offers no effort parameter). **Warn when
   `result.resolved.clerk.model` is anything but `sonnet` or `haiku`** (that
   literal pair is the rule, so it is never re-derived; an unknown custom id also
   warns, with "verify it is cheap"): the clerk transcribes approved
@@ -79,7 +79,7 @@ and report each item's `state` from `result.items[]` — the bin wraps the plan
   and point at `configure` to clean them up — do not present them as the active
   configuration, because they are not.
 
-## `configure` — model per invocation, recorded in config (ADR-008)
+## `configure` — model and effort per invocation, recorded in config (ADR-008; the effort ADR)
 
 > **Why there are no override copies here.** ADR-003 wrote
 > `<project>/.claude/agents/<name>.md` believing an equal `name:` shadows the
@@ -97,14 +97,15 @@ and report each item's `state` from `result.items[]` — the bin wraps the plan
    effort. The one exception is the clerk, which only transcribes approved
    content — it stays cheap regardless of the preset."* Options: keep bundled default (`opus`) / `fable` / `sonnet` /
    custom model ID (free-form). Offer the current session's model as a hint
-   option — you know what you are running on. **Do not ask about effort** — see
-   step 5. **`inherit` is no longer offered**: it meant "follow the session's
+   option — you know what you are running on. Then ask the effort for the
+   preset — `low` / `medium` / `high` / `xhigh` / `max`, or "leave it to the
+   frontmatter" (`max`) — see step 5. **`inherit` is no longer offered**: it meant "follow the session's
    model", and that cannot be expressed per invocation — passing nothing falls
    through to the bundled `model: opus`, not to the session. A user who wants
    session-follow behaviour should pick their session's model explicitly, or set
    `CLAUDE_CODE_SUBAGENT_MODEL=inherit`, which does mean exactly that.
-2. **Optional follow-up**: "configure individually?" → per-agent model for each
-   roster agent. Skippable.
+2. **Optional follow-up**: "configure individually?" → per-agent model and
+   effort for each roster agent. Skippable.
 3. **Apply**: after the AskUserQuestion, run the verb and print its output —
    `node "${CLAUDE_PLUGIN_ROOT}/bin/projectstore.mjs" agents configure --harness claude-code --default <model> [--agent <name>=<model> …] --project "${CLAUDE_PROJECT_DIR}"`
    (naming the harness is the confirmation; the verb writes
@@ -120,24 +121,31 @@ and report each item's `state` from `result.items[]` — the bin wraps the plan
    the emptied block. A name outside the layout's roster is a usage error
    naming the roster — a model written under a name no agent carries would
    never run. That file is the whole output of this command — **never write
-   an agent copy into `.claude/agents/`**. The verb never writes an `effort`
-   key; one already inside the agents block is dropped by the next `configure`
-   write and named in its preview, and until then doctor reports it as a key the
-   overlay may not carry — it has no effect (the effort you configured is not
-   the effort that runs).
+   an agent copy into `.claude/agents/`**. `agents configure` keeps every
+   effort the overlay holds; `--reset` drops them with the models.
+   **Efforts** chosen in steps 1–2 are written by `configure`, the one verb
+   with effort keys — preview first, then apply after the same approval:
+   `node "${CLAUDE_PLUGIN_ROOT}/bin/projectstore.mjs" configure --set agents.default.effort=<level> [--set agents.<name>.effort=<level> …] --harness claude-code --dry-run --project "${CLAUDE_PROJECT_DIR}"`,
+   then the same command without `--dry-run`. `-` as the level clears one. A
+   default effort with no clerk effort named pins the clerk's effort to `low`
+   and says so, as a default model pins its model: the clerk transcribes.
 4. **Migrate away from copies**: if `.claude/agents/` holds copies carrying
    `# source: projectstore v…`, they are pre-ADR-008 leftovers that override
    nothing. Offer to delete them **one approval per file** (matching `/projectstore:doctor --fix`), project scope only — a copy in `~/.claude/agents/` needs a manual removal, and you should say so rather than implying this command will handle it.
    Copies WITHOUT the provenance marker are user-authored — never touch them,
    never mention deleting them.
-5. **Effort is not configurable per project.** The bundled agents ship
-   `effort: max`, which is the recommended value, and there is no
-   per-invocation effort parameter — only frontmatter, settings, or
-   `CLAUDE_CODE_EFFORT_LEVEL`. If the user asks for a different effort, say
-   that plainly and point at the env var; do not write a copy to achieve it.
+5. **Effort rides the spawn, like the model.** The overlay's
+   `agents.per_agent.<name>.effort ?? agents.default.effort` is passed as the
+   spawn's effort parameter; with none, the bundled agents run at their
+   frontmatter's `effort: max`, and the clerk at its `effort: low`. A level outside `low`, `medium`, `high`,
+   `xhigh`, `max` is written with a warning and refused by the spawn. Do not
+   write a copy to achieve an effort.
 6. **Honesty notes to print**: an org `availableModels` allowlist silently
    downgrades excluded models; the `CLAUDE_CODE_SUBAGENT_MODEL` env var
-   overrides everything configured here, per-invocation parameter included.
+   overrides everything configured here, per-invocation parameter included;
+   `CLAUDE_CODE_EFFORT_LEVEL` overrides every configured effort, and a
+   `maxEffortLevel` setting or an organization's effort cap still limits the
+   level an agent runs at.
    `/projectstore:doctor` validates config shape and reports leftover copies —
    not entitlement, and not whether a given spawn actually passed the model.
 7. **No restart is needed** — nothing about the agent list changed. The model
@@ -154,16 +162,17 @@ node "${CLAUDE_PLUGIN_ROOT}/bin/projectstore.mjs" agents model <name> --json --p
 ```
 
 and passes `result.model` as the spawn's model parameter — `null` means pass
-nothing, the agent's own frontmatter decides. The verb applies
-`agents.per_agent.<name>.model ?? agents.default.model ?? null` over
+nothing, the agent's own frontmatter decides — and `result.effort` as its
+effort parameter when it is not `null` and the spawn offers one. The verb applies
+`agents.per_agent.<name>.model ?? agents.default.model ?? null` (and the same
+two terms for `effort`) over
 `<project>/.projectstore/harness/<harness>.json` (the active harness's
 overlay); besides the agents block itself (harness-neutral prose with no bin
 to call — it states the rule as a file read, deliberately), nothing else
 restates that rule. Never guess a model.
 
 `agents.default` is optional and often absent (a per-agent-only config is normal);
-the resolution must tolerate that. An `effort` key, if present, is a pre-ADR-008
-leftover: ignore it.
+the resolution must tolerate that, and so must an effort set without a model.
 
 **Coverage, stated honestly.** This reaches spawns made by this plugin's commands
 and spawns a session makes while following the registration block. It does *not*
@@ -174,7 +183,10 @@ path, at the cost of applying to every subagent on the machine.
 
 ## Notes
 
-- Uninstalling the plugin removes these commands but NOT the block — run
+- `/plugin uninstall` removes these commands but NOT the block — run
   `unregister` first, or delete everything between the
-  `<!-- projectstore:agents -->` markers by hand.
+  `<!-- projectstore:agents -->` markers by hand. From a terminal in the
+  project, the package's bare `uninstall` removes the block too. A one-harness
+  `--harness` run keeps a block in `AGENTS.md` that another harness still
+  reads; `uninstall --surface agents_block` removes it outright.
 - Never write any file without AskUserQuestion approval.

package/node_modules/projectstore/commands/bind.md

@@ -111,7 +111,7 @@ Steps:
 
 8. **Agent registration** (v0.13, ADR-002): ask via AskUserQuestion — "Register projectstore's agents in CLAUDE.md/AGENTS.md so every session routes to them (critic after authoring artifacts, planner before implementing, reviewer before commit)? [Yes (Recommended) / No]". On Yes, run `node "${CLAUDE_PLUGIN_ROOT}/bin/projectstore.mjs" install --harness claude-code --surface agents_block --project "${CLAUDE_PROJECT_DIR}"` and print its output (it renders from the layout's roster, migrates rather than duplicates, previews every write, and applies because the harness is named). On a rebind where a block already exists the verb reports it current or replaces it in place — do not re-ask blindly.
 
-9. **Agent model preset** (v0.13, ADR-003; mechanism per ADR-008): ask via AskUserQuestion — include this line in the question text: *"These agents don't write code — they are critics, planners, and reviewers; they perform best on strong models at high effort."* Options: **Keep bundled default — opus** (Recommended) / **fable** / **sonnet**. Do not offer `inherit` — it cannot be expressed per invocation (see `commands/agents.md`). Do not offer effort — it is not configurable per project (the bundled agents already run at `max`). Free-form model IDs and per-agent tuning live in `/projectstore:agents configure` — mention it. A non-default choice runs the `configure` apply flow from `commands/agents.md`, which writes the config only — **never an agent copy**. Skippable.
+9. **Agent model preset** (v0.13, ADR-003; mechanism per ADR-008): ask via AskUserQuestion — include this line in the question text: *"These agents don't write code — they are critics, planners, and reviewers; they perform best on strong models at high effort."* Options: **Keep bundled default — opus** (Recommended) / **fable** / **sonnet**. Do not offer `inherit` — it cannot be expressed per invocation (see `commands/agents.md`). Do not ask effort here — until one is set, the bundled agents run at their frontmatter's `max`, and the clerk at `low`. Free-form model IDs, per-agent tuning and each agent's effort live in `/projectstore:agents configure` — mention it. A non-default choice runs the `configure` apply flow from `commands/agents.md`, which writes the config only — **never an agent copy**. Skippable.
 
 10. **Print summary**: confirm the bind, list the layout's folders, suggest next commands (`/projectstore:status`, `/projectstore:adr "<first decision>"`, `/projectstore:epic <ID> "<title>"`).
 

package/node_modules/projectstore/commands/doctor.md

@@ -106,13 +106,21 @@ You are running projectstore diagnostics (ADR-005: umbrella doctor).
      verbatim, run from a terminal outside this session.
    - `overlay-forbidden-key` / `overlay-unparseable` (issue) →
      `.projectstore/harness/<id>.json` carries a key an overlay may not (only
-     `agents.default.model` and `agents.per_agent.<name>.model` are read) or is
+     `agents.default.model`, `agents.default.effort`,
+     `agents.per_agent.<name>.model` and `agents.per_agent.<name>.effort` are
+     read) or is
      not JSON. Print the finding. A key inside the agents block goes away on the
      next `/projectstore:agents configure` write; a key outside it, and a parse
      error, are the user's to edit — never rewrite the file yourself.
    - `overlay-unknown-agent` (warn) → the overlay configures a name no roster
      agent carries (a typo, or a newer package's agent): nothing runs under it.
-     Point at `/projectstore:agents configure` with the roster's names.
+     Point at `/projectstore:agents configure` with the roster's names; a name
+     outside the roster is refused by both writers, so removing one is
+     `agents configure --reset` or the user's own edit of the file.
+   - `overlay-unknown-effort` (warn) → the overlay sets an effort the harness's
+     levels lack (a typo, or a leftover such as `Max`): a spawn may refuse it.
+     Point at `/projectstore:agents configure`, which writes efforts through
+     `configure --set agents.<name>.effort=<level>`; `-` clears one.
    - `plugin-registration` (info) → nothing to repair; it names where the npm
      registration loads from. As an **issue** — stale, or two enabled copies —
      print the finding and relay its command verbatim (the package runner's
@@ -128,7 +136,11 @@ You are running projectstore diagnostics (ADR-005: umbrella doctor).
    - `harness` (info) → nothing to repair; it names what `install` can target.
    - `mcp` → the plugin-bundled `.mcp.json` is missing or does not launch
      `bin/projectstore.mjs mcp`: the install is incomplete — the fix is the
-     host's update path (`/plugin update`), never a hand-written file.
+     update for the channel this copy was installed through (`/plugin update`
+     for a git-marketplace copy; for the npm registration, the package's
+     `upgrade` from a terminal in the project, outside this session — the
+     `plugin-registration` finding spells the command), never a hand-written
+     file.
    - `override-copies` → a copy carrying the provenance marker overrides nothing
      (ADR-008): offer to **delete** it (approval-gated, one prompt per file), and
      say that `/projectstore:agents configure` now records the model in

package/node_modules/projectstore/commands/reconcile.md

@@ -38,8 +38,8 @@ You are reconciling the vault's derived views with their source of truth (frontm
    content is authored by the context-holder, the write ceremony by a clerk").
    When the approved set contains two or more targets, hand steps 6-7 to
    `projectstore:clerk`: pass the exact selector list from step 4's preview and
-   the expectation that doctor ends clean. **Model (ADR-008)**: resolve
-   the model with `node "${CLAUDE_PLUGIN_ROOT}/bin/projectstore.mjs" agents model clerk --json --project "${CLAUDE_PROJECT_DIR}"` and pass `result.model` as the spawn's model parameter (`null` → pass nothing);
+   the expectation that doctor ends clean. **Model and effort (ADR-008)**: resolve
+   them with `node "${CLAUDE_PLUGIN_ROOT}/bin/projectstore.mjs" agents model clerk --json --project "${CLAUDE_PROJECT_DIR}"` and pass `result.model` as the spawn's model parameter (`null` → pass nothing), and `result.effort` as its effort parameter when it is not `null` and the spawn offers one;
    missing key, `inherit`, or unreadable config → pass nothing and let the
    agent's own frontmatter decide; never guess a model. The clerk applies
    through the core exactly as step 6 specifies and reports per target. A single

package/node_modules/projectstore/commands/review.md

@@ -64,7 +64,7 @@ You are running a peer review on a projectstore artifact.
 
    Set the agent description to: `Peer-review of {{kind}} artifact at {{path}}`. Pass it as a foreground task (you need the result to continue).
 
-   **Model (ADR-008)**: resolve it with `node "${CLAUDE_PLUGIN_ROOT}/bin/projectstore.mjs" agents model critic --json --project "${CLAUDE_PROJECT_DIR}"` and pass `result.model` as the spawn's model parameter (`null` → pass nothing). Missing key, `inherit`, or unreadable config → pass nothing and let the agent's own frontmatter decide; never guess a model. This is the only way the configured model reaches the agent — there are no override copies (`/projectstore:agents configure`). When falling back to `oh-my-claudecode:critic` or `general-purpose`, pass the same model.
+   **Model and effort (ADR-008)**: resolve them with `node "${CLAUDE_PLUGIN_ROOT}/bin/projectstore.mjs" agents model critic --json --project "${CLAUDE_PROJECT_DIR}"` and pass `result.model` as the spawn's model parameter (`null` → pass nothing), and `result.effort` as its effort parameter when it is not `null` and the spawn offers one. Missing key, `inherit`, or unreadable config → pass nothing and let the agent's own frontmatter decide; never guess a model. This is the only way the configured model and effort reach the agent — there are no override copies (`/projectstore:agents configure`). When falling back to `oh-my-claudecode:critic` or `general-purpose`, pass the same model and effort.
 
 7. **Show findings**: print the agent's report verbatim. Number is its number.
 

package/node_modules/projectstore/commands/story.md

@@ -80,7 +80,8 @@ the two cannot collide).
      files with different jobs: the scratch is what gets copied to the target;
      the baseline is what `--check` compares against. Handing `--check` the
      scratch makes it report drift on every run.
-   - **Model (ADR-008)**: resolve it with `node "${CLAUDE_PLUGIN_ROOT}/bin/projectstore.mjs" agents model clerk --json --project "${CLAUDE_PROJECT_DIR}"` and pass `result.model` as the spawn's model parameter (`null` → pass nothing).
+   - **Model and effort (ADR-008)**: resolve them with `node "${CLAUDE_PLUGIN_ROOT}/bin/projectstore.mjs" agents model clerk --json --project "${CLAUDE_PROJECT_DIR}"` and pass `result.model` as the spawn's model parameter (`null` → pass nothing),
+     and `result.effort` as its effort parameter when it is not `null` and the spawn offers one.
      Missing key, `inherit`, or unreadable config → pass nothing and let the
      agent's own frontmatter decide; never guess a model.
    - Capture doctor's summary line (`node "${CLAUDE_PLUGIN_ROOT}/bin/projectstore.mjs" doctor

package/node_modules/projectstore/docs/getting-started.md

@@ -97,7 +97,7 @@ After setup, every setting is one command away, from a terminal in your project:
 npx projectstore configure
 ```
 
-At a terminal it opens a settings menu: each agent's model, the status line and the band, the template language, and the vault's spec policy and lifecycle gates. ↑ and ↓ move, → opens, space chooses, ← goes back, and Enter shows one plan; a second Enter writes it. A model is chosen from its harness's list: Claude Code's aliases, or Codex's own models. Without a terminal it prints the settings with their values and the file each one lives in. From a script or an agent, name the change; naming it is the confirmation:
+At a terminal it opens a settings menu: each agent's model and effort, the status line and the band, the template language, and the vault's spec policy and lifecycle gates. ↑ and ↓ move, → opens, space chooses, ← goes back, and Enter shows one plan; a second Enter writes it. A model is chosen from its harness's list: Claude Code's aliases, or Codex's own models. Without a terminal it prints the settings with their values and the file each one lives in. From a script or an agent, name the change; naming it is the confirmation:
 
 ```
 npx projectstore configure --set language=de --set lifecycle_gates=on

package/node_modules/projectstore/docs/harnesses.md

@@ -138,8 +138,9 @@ What is known, and how:
 - **Roles are rendered as orchestration skills.** Codex spawns subagents through
   its collaboration tool, not by loading a plugin's `agents/` directory. The
   six ProjectStore roles therefore ship as namespaced skills that resolve their
-  configured model through the core and ask Codex to spawn the role. No effort
-  is forced: it inherits unless the user has configured a model policy.
+  configured model and effort through the core and ask Codex to spawn the role
+  with them. With no effort configured, a role spawned on a configured model
+  runs at that model's own default level.
 - **Multi-file edits reach the activity log and entry rule when their paths
   are absolute.** Codex's `apply_patch` carries paths inside
   `tool_input.command`; the shared extractor reads every `Add`, `Update`,

package/node_modules/projectstore/harnesses/claude-code.json

@@ -49,7 +49,7 @@
       {
         "env": "CLAUDE_CODE_EFFORT_LEVEL",
         "kind": "effort",
-        "beats": "agent frontmatter"
+        "beats": "the spawn's effort parameter and agent frontmatter: with it set, the Agent tool offers no effort parameter at all (measured 2026-10-14, 2.1.296)"
       },
       {
         "env": "CLAUDE_CODE_SUBAGENT_MODEL",
@@ -534,6 +534,45 @@
     "cheap_model": "sonnet",
     "cheap_model_evidence": "measured",
     "cheap_model_reason": "The model the clerk is pinned to when a strong default would otherwise lift it (ADR-008). A literal here, in the one file allowed to name a harness's vocabulary; it used to be a literal in scripts/cli.mjs, which made `agents configure --harness codex` write an Anthropic model name into a Codex overlay. MEASURED in the sense that matters: this is the model the clerk has been running on since ADR-008, in this repository, and tests/overlay.test.mjs asserts the pin end to end.",
+    "effort": {
+      "spawn_param": "effort",
+      "spawn_param_evidence": "measured",
+      "spawn_param_reason": "The Agent tool's `effort` parameter, an enum of the five levels below, in the tool schema Claude Code 2.1.296 gave a session on 2026-10-14 (documented from 2.1.292). A plugin agent with `effort: max` in its frontmatter, spawned with `effort: \"xhigh\"`, ran every request at xhigh; spawned with none, at max (subagent transcripts). An off-enum level fails the tool's schema and no subagent starts. So the overlay's effort is passed here, as the model is (the ADR \"Agent effort is passed at spawn beside the model, from the overlay, in each harness's own levels\", decision 2).",
+      "levels": [
+        {
+          "id": "low",
+          "description": "The fastest and cheapest: little thinking before an answer."
+        },
+        {
+          "id": "medium",
+          "description": "Balances speed and depth: thinks where the task needs it."
+        },
+        {
+          "id": "high",
+          "description": "Thinks longer before answering, for harder work."
+        },
+        {
+          "id": "xhigh",
+          "description": "Extended thinking for hard reviews, plans and designs."
+        },
+        {
+          "id": "max",
+          "description": "The most thinking Claude Code allows: the level the bundled agents' frontmatter names, the clerk's excepted."
+        }
+      ],
+      "levels_reason": "The Agent tool's enum, measured. Claude Code lowers a level a model does not support to the highest one it does (documented), so one list serves every alias and no catalog is read.",
+      "strict": true,
+      "strict_reason": "The spawn takes these levels and nothing else: an off-enum level (ultra) failed the Agent tool's schema and no subagent started (measured 2026-10-14, 2.1.296). So a level off this list is written with a note that the spawn refuses it.",
+      "unset": {
+        "value": "frontmatter",
+        "row": "use frontmatter",
+        "model_default": false,
+        "reason": "With no effort passed, the agent runs at its frontmatter's effort — max for every bundled agent but the clerk, which runs at low — whatever model is passed (transcripts of 2026-10-11/12)."
+      },
+      "cheap": "low",
+      "cheap_evidence": "chosen",
+      "cheap_reason": "The clerk's level, in its frontmatter and as its pin under a default effort (the effort ADR, decision 6): it copies an approved file and runs a pinned ceremony. Chosen by the maintainer on 2026-10-14 (\"клерка да на лоу\"), as on Codex."
+    },
     "models": {
       "spawn_takes": "alias",
       "spawn_takes_evidence": "measured",

package/node_modules/projectstore/harnesses/codex.json

@@ -339,25 +339,61 @@
   "update_instructions": "From a terminal in the project, run `npx projectstore@<version> upgrade --harness codex` and restart Codex afterwards, so the new skills and hooks are loaded. The run fetches the shell at that version, atomically replaces its stable source, and `codex plugin add projectstore@projectstore-npx` materialises the new version; this was measured across two dev builds on codex-cli 0.153.4.",
   "agent_translation": {
     "carries_model": false,
-    "carries_model_reason": "The surface projectstore ships — a plugin's agents/ entry — carries no model on Codex: surfaces.agents is supported:false because no route from a plugin's agents/ directory to collaborationspawn_agent has been found (measured). Codex DOES have a definition file that carries model and effort, recorded below, but it is project- or user-scoped and not plugin-shippable, so contract 12 holds as written: what we ship cannot carry a model. Was true here until 2026-09-08, which contradicted both contract 12 and this file's own agents row.",
+    "carries_model_reason": "The surface projectstore ships — a plugin's agents/ entry — carries no model on Codex: surfaces.agents is supported:false because no route from a plugin's agents/ directory to collaborationspawn_agent has been found (measured). Codex DOES have a definition file that carries model and effort, recorded below, but it is project- or user-scoped and not plugin-shippable, so contract 12 holds as written: what we ship cannot carry a model. Was true here until 2026-09-08, which contradicted both contract 12 and this file's own agents row. What the plugin's role skills pass instead is a value at spawn: `spawn_agent` takes `model` and `reasoning_effort` (measured 2026-10-14, effort block below).",
     "definition_file": ".codex/agents/<name>.toml",
     "definition_file_evidence": "documented",
-    "definition_file_reason": "Codex agent definitions are TOML under .codex/agents/ (project) or ~/.codex/agents/ (personal) and accept `model`, `model_reasoning_effort`, `sandbox_mode` and `mcp_servers`. Documented at learn.chatgpt.com/docs/agent-configuration/subagents, read 2026-09-08; NOT measured, and not reachable from a plugin today. This is the route the rendering story (Per-agent effort returns, PS-AGENTS) has to decide on, and B4 has to measure.",
+    "definition_file_reason": "Codex agent definitions are TOML under .codex/agents/ (project) or ~/.codex/agents/ (personal) and accept `model`, `model_reasoning_effort`, `sandbox_mode` and `mcp_servers`. Documented at learn.chatgpt.com/docs/agent-configuration/subagents, read 2026-09-08; NOT measured, and not reachable from a plugin. Not used: the spawn carries model and effort, so no definition is rendered (the ADR \"Agent effort is passed at spawn beside the model, from the overlay, in each harness's own levels\", alternative A).",
     "cheap_model": "gpt-5.6-luna",
     "cheap_model_evidence": "chosen",
     "cheap_model_reason": "CHOSEN, not measured: the verified plugin session exercised the rendered skill surface, not this optional model pin. It reaches a user's overlay — `agents configure --harness codex` writes it as the clerk's pin — so if the id is wrong the failure is at spawn time, on the harness, not here. The transcriber's model. gpt-5.6-luna at $0.20/$1.20 per Mtok is the cost tier gpt-5.6-sol ($4/$20) is twenty times above, and the clerk composes nothing: it copies an approved file and runs a pinned ceremony (ADR-008). Chosen 2026-09-08 from the published model list, by the same reasoning that pins the clerk to the cheap tier on Claude Code.",
-    "effort_levels": [
-      "low",
-      "medium",
-      "high",
-      "xhigh",
-      "max",
-      "ultra"
-    ],
-    "effort_key": "model_reasoning_effort",
-    "effort_evidence": "documented",
-    "effort_precedence": "explicit spawn value, then the [agents] default from config, then the parent agent's value. A model chosen without an effort uses that model's default effort.",
-    "effort_reason": "Recorded, not yet consumed: the overlay's allowlist rejects an effort key and rewrites it out, deliberately, until the story Per-agent effort returns decides whether a role's effort travels in a rendered definition or at spawn. Documented 2026-09-08; the `ultra` level is Codex's and is not a level Claude Code accepts.",
+    "effort": {
+      "spawn_param": "reasoning_effort",
+      "spawn_param_evidence": "measured",
+      "spawn_param_reason": "`collaboration.spawn_agent` takes `reasoning_effort` beside `model`. Measured on codex-cli 0.162.1, 2026-10-14: a parent on gpt-5.6-sol at xhigh spawned gpt-5.6-luna at low, and the child's first turn_context reads gpt-5.6-luna, low. An effort with no model ran the parent's model at that effort.",
+      "levels": [
+        {
+          "id": "low",
+          "description": "Fast responses with lighter reasoning."
+        },
+        {
+          "id": "medium",
+          "description": "Balances speed and reasoning depth for everyday tasks."
+        },
+        {
+          "id": "high",
+          "description": "Greater reasoning depth for complex problems."
+        },
+        {
+          "id": "xhigh",
+          "description": "Extra high reasoning depth for complex problems."
+        },
+        {
+          "id": "max",
+          "description": "Maximum reasoning depth for the hardest problems."
+        },
+        {
+          "id": "ultra",
+          "description": "Maximum reasoning with automatic task delegation; not every model takes it."
+        }
+      ],
+      "levels_reason": "The floor, offered when the agent's model is not in a list that names its own levels: every level Codex's catalog listed on 2026-10-14, in its own words. Not every model takes every level (gpt-5.6-luna stops at max), which is why a listed model's own levels win.",
+      "catalog": {
+        "list": "supported_reasoning_levels",
+        "id": "effort",
+        "description": "description",
+        "default": "default_reasoning_level",
+        "reason": "Fields of a model entry in Codex's own catalog (models_cache.json, `codex debug models`), measured 2026-10-14 on codex-cli 0.162.1: each model lists the levels it takes as { effort, description } and names its default."
+      },
+      "unset": {
+        "value": "Codex decides",
+        "row": "let Codex decide",
+        "model_default": true,
+        "reason": "With no effort passed and a model passed, the child runs at that model's own default level, not the parent's: gpt-5.6-sol ran at low under an xhigh parent (measured 2026-10-14; Codex's subagents documentation says the same). With no model passed either, Codex's [agents] default, then the parent's effort (documented)."
+      },
+      "cheap": "low",
+      "cheap_evidence": "chosen",
+      "cheap_reason": "The clerk's level under a default effort (the ADR \"Agent effort is passed at spawn beside the model, from the overlay, in each harness's own levels\", decision 6): it copies an approved file and runs a pinned ceremony. Chosen by the maintainer on 2026-09-08 with its model: gpt-5.6-luna at the lowest level."
+    },
     "models": {
       "spawn_takes": "id",
       "spawn_takes_evidence": "documented",

package/node_modules/projectstore/package.json

@@ -1,6 +1,6 @@
 {
   "name": "projectstore",
-  "version": "0.30.2",
+  "version": "0.30.3",
   "description": "Your agent runs the project through a verified loop: task → artifact (ADR / spec / epic / story) → adversarial critic → backlog → planner → reviewer → done. Plain markdown in git — any model can pick the project up tomorrow.",
   "keywords": [
     "project-management",

package/node_modules/projectstore/scripts/agents-change.mjs

@@ -7,9 +7,20 @@ import { harnessForOverlay } from "./harness.mjs";
 
 // What a change would write. `before` is readOverlayAt's read; `roster` the
 // layout's agent names (null when no layout loads); `agents` the
-// `<name>=<model>` specs. A malformed spec or a name outside the roster comes
-// back as `usage`, with nothing planned.
-export function planAgentsChange({ before, harness, layout, roster, defaultModel, agents = [], reset = false }) {
+// `<name>=<model>` specs; `defaultEffort` and `efforts` (`<name>=<level>`)
+// the same for effort (the effort ADR, decision 8). A malformed spec or a name
+// outside the roster comes back as `usage`, with nothing planned.
+export function planAgentsChange({
+  before,
+  harness,
+  layout,
+  roster,
+  defaultModel,
+  agents = [],
+  defaultEffort,
+  efforts = [],
+  reset = false,
+}) {
   // --reset empties the block first; --default and --agent then apply on top of
   // the emptied block — "reset, then set" is one call, and nothing given is
   // silently dropped.
@@ -27,6 +38,22 @@ export function planAgentsChange({ before, harness, layout, roster, defaultModel
     if (m[2]) next.per_agent[m[1]] = m[2];
     else delete next.per_agent[m[1]];
   }
+  // Efforts, field by field beside the models: a run that names none keeps
+  // every effort the overlay holds, and --reset empties them with the models.
+  const beforeEfforts = before.efforts || { default: null, per_agent: {} };
+  const baseEfforts = reset ? { default: null, per_agent: {} } : beforeEfforts;
+  const nextEfforts = {
+    default: defaultEffort !== undefined ? defaultEffort || null : baseEfforts.default,
+    per_agent: { ...baseEfforts.per_agent },
+  };
+  for (const spec of efforts) {
+    const m = /^([a-z][a-z0-9-]*)=(.*)$/.exec(spec);
+    if (!m) return { usage: `an effort takes <name>=<level>, got ${spec}` };
+    if (roster && !roster.includes(m[1]))
+      return { usage: `no agent named ${m[1]} in the ${layout} roster — known: ${roster.join(", ")}` };
+    if (m[2]) nextEfforts.per_agent[m[1]] = m[2];
+    else delete nextEfforts.per_agent[m[1]];
+  }
   // The clerk transcribes; a strong default must not lift it (commands/agents.md,
   // ADR-008). WHICH model it is pinned to is the harness's vocabulary, so the
   // manifest names it and this file does not: pinning every harness to
@@ -49,12 +76,29 @@ export function planAgentsChange({ before, harness, layout, roster, defaultModel
       pinnedClerk = true;
     } else clerkPinUnavailable = true;
   }
-  // Keys inside the agents block that the allowlist rejected (an `effort`, a
-  // stray shape) are rewritten out by the write and announced here; keys
-  // outside the block are kept, and stay doctor's to name.
+  // The clerk's effort, pinned the same way while a default effort stands
+  // (the effort ADR, decision 6), where the harness names a cheap level; a
+  // harness that names none pins nothing and says nothing, because its clerk
+  // already runs at its frontmatter's level.
+  let pinnedClerkEffort = false;
+  const clerkEffortAsked = efforts.some((s) => s.startsWith("clerk="));
+  if (nextEfforts.default && !nextEfforts.per_agent.clerk && !clerkEffortAsked) {
+    const cheap = harnessForOverlay(harness)?.agent_translation?.effort?.cheap || null;
+    if (cheap) {
+      nextEfforts.per_agent.clerk = cheap;
+      pinnedClerkEffort = true;
+    }
+  }
+  // Keys inside the agents block that the allowlist rejected (a stray shape)
+  // are rewritten out by the write and announced here; keys outside the block
+  // are kept, and stay doctor's to name.
   const dropped = before.rejected.filter((k) => k.startsWith("agents."));
+  const sameEfforts =
+    JSON.stringify({ d: beforeEfforts.default, p: beforeEfforts.per_agent }) ===
+    JSON.stringify({ d: nextEfforts.default, p: nextEfforts.per_agent });
   const same =
     !dropped.length &&
+    sameEfforts &&
     JSON.stringify({ d: before.agents.default, p: before.agents.per_agent }) ===
       JSON.stringify({ d: next.default, p: next.per_agent });
   const preview = [
@@ -72,13 +116,23 @@ export function planAgentsChange({ before, harness, layout, roster, defaultModel
     preview.push(
       `  clerk: takes the default — ${harness} declares no cheap model to pin it to, so name one with --agent clerk=<model>`,
     );
+  // Effort lines only where an effort is held or asked for, so a run over an
+  // overlay of models alone previews exactly as it did before efforts.
+  if (beforeEfforts.default || nextEfforts.default)
+    preview.push(`  default effort: ${beforeEfforts.default || "—"} → ${nextEfforts.default || "—"}`);
+  const effortNames = new Set([...Object.keys(beforeEfforts.per_agent), ...Object.keys(nextEfforts.per_agent)]);
+  for (const n of [...effortNames].sort())
+    preview.push(
+      `  ${n} effort: ${beforeEfforts.per_agent[n] || "—"} → ${nextEfforts.per_agent[n] || "—"}${pinnedClerkEffort && n === "clerk" ? " (pinned: the clerk stays cheap under a strong default)" : ""}`,
+    );
   if (dropped.length)
-    preview.push(`  drops from the agents block (an overlay carries only models): ${dropped.join(", ")}`);
-  return { next, pinnedClerk, clerkPinUnavailable, dropped, same, preview };
+    preview.push(`  drops from the agents block (an overlay carries only models and efforts): ${dropped.join(", ")}`);
+  return { next, nextEfforts, pinnedClerk, pinnedClerkEffort, clerkPinUnavailable, dropped, same, preview };
 }
 
 // Writes a planned change through the overlay's one writer and reads it back.
-export function applyAgentsChange(projectDir, harness, next) {
-  const path = writeOverlayAt(projectDir, harness, next);
+// `nextEfforts` left out keeps the efforts the file holds.
+export function applyAgentsChange(projectDir, harness, next, nextEfforts) {
+  const path = writeOverlayAt(projectDir, harness, next, nextEfforts);
   return { path, after: readOverlayAt(projectDir, harness) };
 }