@vibe-hero/server 0.1.0

package/LICENSE

@@ -0,0 +1,190 @@
+                                 Apache License
+                           Version 2.0, January 2004
+                        http://www.apache.org/licenses/
+
+   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
+
+   1. Definitions.
+
+      "License" shall mean the terms and conditions for use, reproduction,
+      and distribution as defined by Sections 1 through 9 of this document.
+
+      "Licensor" shall mean the copyright owner or entity authorized by
+      the copyright owner that is granting the License.
+
+      "Legal Entity" shall mean the union of the acting entity and all
+      other entities that control, are controlled by, or are under common
+      control with that entity. For the purposes of this definition,
+      "control" means (i) the power, direct or indirect, to cause the
+      direction or management of such entity, whether by contract or
+      otherwise, or (ii) ownership of fifty percent (50%) or more of the
+      outstanding shares, or (iii) beneficial ownership of such entity.
+
+      "You" (or "Your") shall mean an individual or Legal Entity
+      exercising permissions granted by this License.
+
+      "Source" form shall mean the preferred form for making modifications,
+      including but not limited to software source code, documentation
+      source, and configuration files.
+
+      "Object" form shall mean any form resulting from mechanical
+      transformation or translation of a Source form, including but
+      not limited to compiled object code, generated documentation,
+      and conversions to other media types.
+
+      "Work" shall mean the work of authorship, whether in Source or
+      Object form, made available under the License, as indicated by a
+      copyright notice that is included in or attached to the work
+      (an example is provided in the Appendix below).
+
+      "Derivative Works" shall mean any work, whether in Source or Object
+      form, that is based on (or derived from) the Work and for which the
+      editorial revisions, annotations, elaborations, or other modifications
+      represent, as a whole, an original work of authorship. For the purposes
+      of this License, Derivative Works shall not include works that remain
+      separable from, or merely link (or bind by name) to the interfaces of,
+      the Work and Derivative Works thereof.
+
+      "Contribution" shall mean any work of authorship, including
+      the original version of the Work and any modifications or additions
+      to that Work or Derivative Works thereof, that is intentionally
+      submitted to the Licensor for inclusion in the Work by the copyright owner
+      or by an individual or Legal Entity authorized to submit on behalf of
+      the copyright owner. For the purposes of this definition, "submitted"
+      means any form of electronic, verbal, or written communication sent
+      to the Licensor or its representatives, including but not limited to
+      communication on electronic mailing lists, source code control systems,
+      and issue tracking systems that are managed by, or on behalf of, the
+      Licensor for the purpose of discussing and improving the Work, but
+      excluding communication that is conspicuously marked or otherwise
+      designated in writing by the copyright owner as "Not a Contribution."
+
+      "Contributor" shall mean Licensor and any individual or Legal Entity
+      on behalf of whom a Contribution has been received by the Licensor and
+      subsequently incorporated within the Work.
+
+   2. Grant of Copyright License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      copyright license to reproduce, prepare Derivative Works of,
+      publicly display, publicly perform, sublicense, and distribute the
+      Work and such Derivative Works in Source or Object form.
+
+   3. Grant of Patent License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      (except as stated in this section) patent license to make, have made,
+      use, offer to sell, sell, import, and otherwise transfer the Work,
+      where such license applies only to those patent claims licensable
+      by such Contributor that are necessarily infringed by their
+      Contribution(s) alone or by combination of their Contribution(s)
+      with the Work to which such Contribution(s) was submitted. If You
+      institute patent litigation against any entity (including a
+      cross-claim or counterclaim in a lawsuit) alleging that the Work
+      or a Contribution incorporated within the Work constitutes direct
+      or contributory patent infringement, then any patent licenses
+      granted to You under this License for that Work shall terminate
+      as of the date such litigation is filed.
+
+   4. Redistribution. You may reproduce and distribute copies of the
+      Work or Derivative Works thereof in any medium, with or without
+      modifications, and in Source or Object form, provided that You
+      meet the following conditions:
+
+      (a) You must give any other recipients of the Work or
+          Derivative Works a copy of this License; and
+
+      (b) You must cause any modified files to carry prominent notices
+          stating that You changed the files; and
+
+      (c) You must retain, in the Source form of any Derivative Works
+          that You distribute, all copyright, patent, trademark, and
+          attribution notices from the Source form of the Work,
+          excluding those notices that do not pertain to any part of
+          the Derivative Works; and
+
+      (d) If the Work includes a "NOTICE" text file as part of its
+          distribution, then any Derivative Works that You distribute must
+          include a readable copy of the attribution notices contained
+          within such NOTICE file, excluding any notices that do not
+          pertain to any part of the Derivative Works, in at least one
+          of the following places: within a NOTICE text file distributed
+          as part of the Derivative Works; within the Source form or
+          documentation, if provided along with the Derivative Works; or,
+          within a display generated by the Derivative Works, if and
+          wherever such third-party notices normally appear. The contents
+          of the NOTICE file are for informational purposes only and
+          do not modify the License. You may add Your own attribution
+          notices within Derivative Works that You distribute, alongside
+          or as an addendum to the NOTICE text from the Work, provided
+          that such additional attribution notices cannot be construed
+          as modifying the License.
+
+      You may add Your own copyright statement to Your modifications and
+      may provide additional or different license terms and conditions
+      for use, reproduction, or distribution of Your modifications, or
+      for any such Derivative Works as a whole, provided Your use,
+      reproduction, and distribution of the Work otherwise complies with
+      the conditions stated in this License.
+
+   5. Submission of Contributions. Unless You explicitly state otherwise,
+      any Contribution intentionally submitted for inclusion in the Work
+      by You to the Licensor shall be under the terms and conditions of
+      this License, without any additional terms or conditions.
+      Notwithstanding the above, nothing herein shall supersede or modify
+      the terms of any separate license agreement you may have executed
+      with Licensor regarding such Contributions.
+
+   6. Trademarks. This License does not grant permission to use the trade
+      names, trademarks, service marks, or product names of the Licensor,
+      except as required for reasonable and customary use in describing the
+      origin of the Work and reproducing the content of the NOTICE file.
+
+   7. Disclaimer of Warranty. Unless required by applicable law or
+      agreed to in writing, Licensor provides the Work (and each
+      Contributor provides its Contributions) on an "AS IS" BASIS,
+      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+      implied, including, without limitation, any warranties or conditions
+      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+      PARTICULAR PURPOSE. You are solely responsible for determining the
+      appropriateness of using or redistributing the Work and assume any
+      risks associated with Your exercise of permissions under this License.
+
+   8. Limitation of Liability. In no event and under no legal theory,
+      whether in tort (including negligence), contract, or otherwise,
+      unless required by applicable law (such as deliberate and grossly
+      negligent acts) or agreed to in writing, shall any Contributor be
+      liable to You for damages, including any direct, indirect, special,
+      incidental, or consequential damages of any character arising as a
+      result of this License or out of the use or inability to use the
+      Work (including but not limited to damages for loss of goodwill,
+      work stoppage, computer failure or malfunction, or any and all
+      other commercial damages or losses), even if such Contributor
+      has been advised of the possibility of such damages.
+
+   9. Accepting Warranty or Additional Liability. While redistributing
+      the Work or Derivative Works thereof, You may choose to offer,
+      and charge a fee for, acceptance of support, warranty, indemnity,
+      or other liability obligations and/or rights consistent with this
+      License. However, in accepting such obligations, You may act only
+      on Your own behalf and on Your sole responsibility, not on behalf
+      of any other Contributor, and only if You agree to indemnify,
+      defend, and hold each Contributor harmless for any liability
+      incurred by, or claims asserted against, such Contributor by reason
+      of your accepting any such warranty or additional liability.
+
+   END OF TERMS AND CONDITIONS
+
+   Copyright 2026 Sjors Robroek
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.

package/README.md

@@ -0,0 +1,151 @@
+# @vibe-hero/server
+
+The vibe-hero MCP server. Exposes 10 MCP tools over stdio that a host agent
+(Claude Code, Codex, Kiro) calls to drive adaptive learning sessions: quiz
+delivery, Elo-style ability tracking, tier graduation, telemetry intake, and
+offer management. All learning state lives here — the host model is stateless.
+
+## Requirements
+
+- Node >= 18
+- pnpm (workspace root manages dependencies)
+
+## Build and run
+
+```sh
+# From the repository root — install all workspace deps first
+pnpm install
+
+# Compile the server
+pnpm --filter @vibe-hero/server build
+
+# Run tests
+pnpm --filter @vibe-hero/server test
+
+# Start the MCP server over stdio (for manual inspection / MCP inspector)
+VIBE_HERO_HOME="$(mktemp -d)" pnpm --filter @vibe-hero/server start
+```
+
+`start` executes `node dist/index.js`. The server communicates over stdio using
+the MCP protocol; it is not a long-running HTTP service.
+
+## Environment variables
+
+| Variable | Default | Purpose |
+|---|---|---|
+| `VIBE_HERO_HOME` | `~/.vibe-hero` | Root directory for the user profile (`profile.json`) and the content cache (`content/`). Point at a temp dir during tests to avoid touching a real profile. |
+| `VIBE_HERO_CONTENT_URL` | *(disabled)* | URL of a remote content catalog to fetch and cache. When unset, the server serves only the bundled baseline content that ships with the package. |
+
+## Profile location
+
+The learner profile is stored at `$VIBE_HERO_HOME/profile.json`
+(default `~/.vibe-hero/profile.json`). It records per-(topic × tool) ability
+estimates, tier graduation state, the review schedule, quiz history, and the
+offer ledger. The profile is global per user — it is not per-project.
+
+Profile writes are atomic (write-temp + rename under an advisory lock) so
+concurrent host sessions cannot clobber each other's updates.
+
+## The 10 MCP tools
+
+Every tool except `get_config` and `save_config` returns a
+`{ "status": "SETUP_REQUIRED" }` sentinel when the profile has no
+configuration. The host agent should respond by running the `vibe-hero-setup`
+skill.
+
+| Tool | Description |
+|---|---|
+| `get_status` | Per-topic standing and tier for a tool (or all tools). Read-only; never requires telemetry. |
+| `list_topics` | Enumerate catalog topics, optionally filtered by tool or class. Returns `catalogVersion`. |
+| `get_guidance` | Teaching text and a suggested next step for a topic, or the weakest/stalest area if no topic is named. |
+| `start_quiz` | Begin a quiz session: selects 3–5 items by difficulty-targeting near current ability. Returns `PresentedItem[]` — answer keys are never included for deterministic items; `rubric` + `referenceAnswer` are included for free-form items so the host agent can judge. |
+| `submit_answer` | Grade one item and update the ability estimate. Deterministic types (multiple-choice, short-answer) are graded by the engine; free-form types accept a per-criterion verdict from the host agent and the engine computes the score. Returns grade, guidance, ability delta, and any graduation change. |
+| `save_config` | Persist the configuration produced by the setup Q&A. Clears the setup gate. Safe to re-run; never wipes learning progress. |
+| `get_config` | Read current configuration (or absence). Used by skills and the Stop hook to check gate state. |
+| `record_observation` | Intake for the Stop hook / telemetry source. Maps observed tool activity to candidate topics for offers. Stores only derived signals — never raw tool input/output. Awards nothing (usage never scores). |
+| `get_offer` | Resolve whether to surface an end-of-work offer for the current session. Called by the Stop hook. Returns an offer or a suppression reason. |
+| `record_offer_response` | Record accept / decline / defer so cadence and anti-nag rules are honored. A decline suppresses further offers for the session; repeated cross-session declines increase backoff and eventually mute offers globally. |
+
+Full input/output shapes: `specs/001-vibe-hero-mvp/contracts/mcp-tools.md`.
+
+## Content model
+
+Content is organized as a matrix of (topic × class × tier):
+
+- **Tier** — 100 to 500 on a cognitive-depth scale (100 = introductory, 500 = expert).
+- **Class** — `general` (tool-agnostic agentic-coding concepts) or `tool:<id>`
+  (e.g. `tool:claude-code`). Graduation is tracked independently per tool.
+- **Topic** — one YAML file per (topic × class) under `content/`. Each file
+  contains all tier items for that topic plus trigger-signal declarations (the
+  tool names the engine uses to attribute observed activity to the topic).
+
+Bundled baseline content ships with the package (offline-capable). If
+`VIBE_HERO_CONTENT_URL` is set, the server fetches and caches an updated
+catalog; on failure it falls back to the cached or bundled copy.
+
+v1 content: `content/claude-code/` (subagents, context-management, planning)
+and `content/general/` (task-decomposition).
+
+## Assessment model
+
+- **Ability estimates** are per-(topic × class) and update on every graded
+  answer using an Elo-style formula against fixed item difficulty ratings.
+  Item difficulty is authored and never self-updates.
+- **Graduation** to a tier requires crossing the tier boundary by a margin
+  (hysteresis band); demotion requires crossing a separate lower threshold.
+  Both require holding the crossing for 2 consecutive graded items (dwell),
+  preventing flip-flopping at the boundary.
+- **Lapse / review scheduling** — if ability decays below the demotion
+  threshold or 30 days pass without assessment, the topic is surfaced for
+  review.
+- Tunable parameters (K-factor, tier boundaries, hysteresis margin, decay
+  half-life, etc.) live in `src/config.ts`.
+
+## Privacy
+
+The server never persists raw prompts, tool inputs, tool outputs, or any user
+content. Observation intake extracts only derived signals:
+`{ tool_name, topicKeys, success, timestamp, tool_use_id }`. The profile
+contains only ability estimates, scores, timestamps, and configuration — no
+content the user typed.
+
+## Wiring into Claude Code
+
+### MCP server
+
+Add to `.claude/settings.json` (project) or `~/.claude/settings.json` (global):
+
+```jsonc
+{
+  "mcpServers": {
+    "vibe-hero": {
+      "type": "stdio",
+      "command": "node",
+      "args": ["/absolute/path/to/vibe-hero/packages/server/dist/index.js"]
+    }
+  }
+}
+```
+
+### Stop hook (end-of-work offers)
+
+See `hooks/claude-code/README.md` for how to register `hooks/claude-code/stop-offer.sh`
+as a Claude Code Stop hook. The hook requires `jq` on `PATH` and the server
+built (dist present). It is advisory-only: it exits 0 silently on any error.
+
+### Skills (portable agent surface)
+
+Copy or reference the four skills from `skills/` into your agent skill path:
+
+| Skill | Purpose |
+|---|---|
+| `vibe-hero-setup` | Required first-run Q&A; collects preferences and calls `save_config`. Must run before any other skill works. |
+| `vibe-hero-quiz` | Drives the `start_quiz` / `submit_answer` loop; judges free-form items against the MCP-supplied rubric. |
+| `vibe-hero-status` | Calls `get_status` and renders per-topic standing and suggestions. |
+| `vibe-hero-learn` | Calls `get_guidance` (and optionally `list_topics`) to surface teaching text and a next step. |
+
+## Distribution
+
+npm publishing and Claude Code plugin/marketplace bundling are planned as a
+follow-up (spec 002). Currently the server is set up manually as described
+above.

package/dist/catalog/bundled/claude-code/context-management.yaml

@@ -0,0 +1,302 @@
+# Topic: context-management (claude-code)
+#
+# Covers Claude Code context window hygiene: reading strategically, avoiding
+# unnecessary re-reads, compaction, the TodoWrite/TodoRead pattern, and how
+# large context affects performance and cost. Tiers 100–500.
+
+id: context-management
+class:
+  kind: tool
+  tool: claude-code
+title: Context Window Management
+summary: >-
+  Keeping the context window lean and effective — strategic reads, avoiding
+  re-reads, compaction signals, the Todo list as working memory, and what
+  happens when context fills up.
+
+triggerSignals:
+  - tool: claude-code
+    match:
+      toolName: TodoWrite
+    weight: 1
+  - tool: claude-code
+    match:
+      toolName: TodoRead
+    weight: 0.8
+  - tool: claude-code
+    match:
+      toolNamePattern: "^Read$"
+    weight: 0.5
+  - tool: claude-code
+    match:
+      mcpToolPattern: ".*compact.*"
+    weight: 0.7
+
+items:
+  # ── Tier 100 — Remember ──────────────────────────────────────────────────
+  - id: ctx-100-mc-a
+    tier: 100
+    bloom: remember
+    difficulty: 150
+    type: multiple_choice
+    prompt: >-
+      After successfully writing a file with the Edit or Write tool, should you
+      immediately re-read the file to verify the change?
+    choices:
+      - id: a
+        text: >-
+          Yes — always re-read to confirm the file on disk matches what you
+          intended
+      - id: b
+        text: >-
+          No — Edit/Write would have errored if the change failed; re-reading
+          wastes context
+      - id: c
+        text: >-
+          Only re-read files larger than 100 lines
+      - id: d
+        text: >-
+          Re-read only when using Write, not Edit
+    answerKey:
+      kind: choice
+      correctChoiceId: b
+    guidance: >-
+      The Edit and Write tools error on failure; if they return successfully the
+      change landed. Immediately re-reading the file after a successful edit
+      adds the full file content to context for no benefit — a costly habit at
+      scale.
+
+  - id: ctx-100-mc-b
+    tier: 100
+    bloom: remember
+    difficulty: 160
+    type: multiple_choice
+    prompt: >-
+      What is the TodoWrite tool used for in Claude Code?
+    choices:
+      - id: a
+        text: Writing TODO comments into source code files
+      - id: b
+        text: >-
+          Maintaining an in-session task list that tracks planned, in-progress,
+          and completed work
+      - id: c
+        text: Scheduling background tasks for later sessions
+      - id: d
+        text: Sending task notifications to the user
+    answerKey:
+      kind: choice
+      correctChoiceId: b
+    guidance: >-
+      TodoWrite (paired with TodoRead) is Claude Code's in-session task list.
+      It acts as working memory — externalizing what needs to be done so the
+      model doesn't have to hold the full plan in context. Mark tasks
+      in_progress when starting and completed when done.
+
+  # ── Tier 200 — Understand ───────────────────────────────────────────────
+  - id: ctx-200-mc-a
+    tier: 200
+    bloom: understand
+    difficulty: 250
+    type: multiple_choice
+    prompt: >-
+      You need to understand a function defined in a 2000-line file. What is
+      the most context-efficient approach?
+    choices:
+      - id: a
+        text: Read the entire file first, then locate the function
+      - id: b
+        text: >-
+          Use Read with offset and limit parameters to read only the relevant
+          section, or use Grep/a symbolic tool to find the function first
+      - id: c
+        text: Ask the user to paste the function into the chat
+      - id: d
+        text: Read the first 100 lines, then guess the rest
+    answerKey:
+      kind: choice
+      correctChoiceId: b
+    guidance: >-
+      Reading an entire large file when you only need a section is one of the
+      most common context-wasting patterns. Use Read with `offset`/`limit`, or
+      use Grep/symbolic tools to find the exact location first, then read only
+      that range. The instructions explicitly say to prefer dedicated tools over
+      cat/head/tail for this reason.
+
+  - id: ctx-200-sa-a
+    tier: 200
+    bloom: understand
+    difficulty: 260
+    type: short_answer
+    prompt: >-
+      What happens to a Claude Code session's context when it gets very full,
+      and what is the term for the automatic process that trims it?
+    answerKey:
+      kind: keyword
+      anyOf:
+        - compaction
+        - compact
+        - context compaction
+        - auto-compaction
+      normalize: lower
+    guidance: >-
+      When the context window fills up, Claude Code performs *compaction* —
+      it summarizes the conversation history to free space. Compaction is
+      automatic but lossy: details from earlier in the session may be
+      summarized away. Keeping context lean from the start delays compaction
+      and reduces information loss.
+
+  # ── Tier 300 — Apply ────────────────────────────────────────────────────
+  - id: ctx-300-mc-a
+    tier: 300
+    bloom: apply
+    difficulty: 350
+    type: multiple_choice
+    prompt: >-
+      You are starting a multi-file refactor that will span many steps. Before
+      writing a single line of code, which TodoWrite action is most valuable?
+    choices:
+      - id: a
+        text: >-
+          Write a single "refactor everything" task and mark it in_progress
+      - id: b
+        text: >-
+          Break the work into discrete tasks (each file or logical unit), write
+          them all to the Todo list, then start the first one
+      - id: c
+        text: Skip TodoWrite — it adds tool calls that cost context
+      - id: d
+        text: >-
+          Write all tasks as completed so the user sees the plan immediately
+    answerKey:
+      kind: choice
+      correctChoiceId: b
+    guidance: >-
+      Using TodoWrite to lay out discrete, ordered tasks before starting serves
+      as a lightweight project plan external to context. It lets you mark tasks
+      in_progress and completed as you go, which both keeps you on track and
+      signals to the user that progress is being made. One coarse task gives no
+      recovery point if context is compacted mid-session.
+
+  - id: ctx-300-mc-b
+    tier: 300
+    bloom: apply
+    difficulty: 360
+    type: multiple_choice
+    prompt: >-
+      Your session's context contains the full text of five large files you
+      read earlier but no longer need. You are about to start a complex new
+      subtask. What is the best strategy?
+    choices:
+      - id: a
+        text: >-
+          Continue normally — the model handles large context automatically
+      - id: b
+        text: >-
+          Ask the user to run /compact or /clear before proceeding so stale
+          large-file content is removed from context
+      - id: c
+        text: >-
+          Re-read the files you still need, which signals to the model what is
+          important
+      - id: d
+        text: >-
+          Open a new terminal session and start over
+    answerKey:
+      kind: choice
+      correctChoiceId: b
+    guidance: >-
+      Stale large-file content sitting in context is pure waste — it uses
+      tokens, may mislead later reasoning, and accelerates compaction. For a
+      fresh subtask, prompting the user to run /compact (summarize history) or
+      /clear (reset) is the correct hygiene move. The model does not
+      automatically prune irrelevant earlier reads.
+
+  # ── Tier 400 — Analyze ──────────────────────────────────────────────────
+  - id: ctx-400-mc-a
+    tier: 400
+    bloom: analyze
+    difficulty: 430
+    type: multiple_choice
+    prompt: >-
+      In a long session you notice the model is forgetting instructions given
+      at the start of the conversation. What is the most likely cause?
+    choices:
+      - id: a
+        text: A bug in Claude Code
+      - id: b
+        text: >-
+          Compaction has summarized away the early instructions to free context
+          space
+      - id: c
+        text: The CLAUDE.md file was not loaded
+      - id: d
+        text: The model intentionally ignores earlier turns in long sessions
+    answerKey:
+      kind: choice
+      correctChoiceId: b
+    guidance: >-
+      When context fills and compaction runs, the summary may lose or dilute
+      instructions given early in the session. Key persistent instructions
+      belong in CLAUDE.md (loaded fresh each session) or in the system prompt,
+      not only in early conversation turns. Compaction is expected behavior, not
+      a bug.
+
+  - id: ctx-400-sa-a
+    tier: 400
+    bloom: analyze
+    difficulty: 440
+    type: short_answer
+    prompt: >-
+      You must read a PDF file that is 25 pages long. Which Read tool parameter
+      must you provide (besides file_path) to avoid a failure, and what is the
+      maximum number of pages per request?
+    answerKey:
+      kind: keyword
+      anyOf:
+        - pages
+        - "pages parameter"
+        - "20"
+        - "20 pages"
+      normalize: lower
+    guidance: >-
+      The Read tool requires the `pages` parameter for large PDFs (more than 10
+      pages) and allows a maximum of 20 pages per request. Attempting to read a
+      large PDF without `pages` will fail. For a 25-page document you would need
+      two reads: e.g., pages "1-20" then pages "21-25".
+
+  # ── Tier 500 — Evaluate ─────────────────────────────────────────────────
+  - id: ctx-500-mc-a
+    tier: 500
+    bloom: evaluate
+    difficulty: 480
+    type: multiple_choice
+    prompt: >-
+      You are designing a workflow where Claude Code repeatedly re-reads the
+      same configuration file at the start of each subtask "to make sure it has
+      the latest content." Evaluate this pattern.
+    choices:
+      - id: a
+        text: >-
+          Good practice — always read fresh to avoid acting on stale data
+      - id: b
+        text: >-
+          Acceptable for small files; wasteful for large ones
+      - id: c
+        text: >-
+          Anti-pattern: the file was already read and unchanged reads just
+          inflate context with duplicate content; only re-read when the file
+          might have changed
+      - id: d
+        text: >-
+          Required — the Read tool does not cache content between tool calls
+    answerKey:
+      kind: choice
+      correctChoiceId: c
+    guidance: >-
+      Re-reading a file that has not changed since the last read is pure context
+      inflation. Each re-read copies the full file text into context again,
+      accelerating compaction and increasing cost. The correct pattern is to
+      read once, act on the result, and only re-read if an intervening action
+      (Edit/Write/Bash) has modified the file — or if the file is produced
+      externally and genuinely may have changed.