7 changed files with 163 additions and 390 deletions
--- a/CLAUDE.md
+++ b/CLAUDE.md
@ -10,9 +10,9 @@
 ## Current Project State
- **Phase:** Active development — Phase 1 + 2 + 2.5 complete; Phase 3 (investigation planning) ready to start
+- **Phase:** Active development — Phase 2 (survey pass) and Phase 2.5 (context budget) complete; Phase 3 (investigation planning) ready to start
- **Last worked on:** 2026-04-07
+- **Last worked on:** 2026-04-06
- **Last commit:** merge: fix/issue-54-write-cache-tool-desc
+- **Last commit:** merge: feat/issue-44-context-budget (#44)
 - **Blocking:** None
 ---
@ -79,15 +79,42 @@ python3 luminos.py --install-extras
 ---
-## Project-Specific Test Notes
+## Development Workflow
-Run tests with `python3 -m unittest discover -s tests/`. Modules exempt from
+- **Issue-driven work** — all work must be tied to a Forgejo issue. If the
-unit testing: `ai.py` (requires live API), `ast_parser.py` (requires
+  user names a specific issue, use it. If they describe work without an issue
-tree-sitter), `watch.py` (stateful events), `prompts.py` (string templates
+  number, search open issues for a match. If no issue exists, gather enough
-only).
+  context to create one before starting work. Branches and commits should
  reference the issue number.
 - **Explain then build** — articulate the approach in a few bullets before
  writing code. Surface assumptions early.
 - **Atomic commits** — each commit is one logical change.
 - **Test coverage required** — every change to a testable module must include
  or update tests in `tests/`. Run with `python3 -m unittest discover -s tests/`.
  All tests must pass before merging. Modules exempt from unit testing:
  `ai.py` (requires live API), `ast_parser.py` (requires tree-sitter),
  `watch.py` (stateful events), `prompts.py` (string templates only).
 - **Shiny object capture** — new ideas go to PLAN.md (Raw Thoughts) or a
  Forgejo issue, not into current work.
-(Development workflow, branching discipline, and session protocols live in
+---
-`~/.claude/CLAUDE.md`.)
+
 ## Branching Discipline
 - **Always branch** — no direct commits to main, ever
 - **Branch before first change** — create the branch before touching any files
 - **Naming:** `feat/`, `fix/`, `refactor/`, `chore/` + short description
 - **One branch, one concern** — don't mix unrelated changes
 - **Two-branch maximum** — never have more than 2 unmerged branches
 - **Merge with `--no-ff`** — preserves branch history in the log
 - **Delete after merge** — `git branch -d <branch>` immediately after merge
 - **Close the underlying issue manually** — after merging, `PATCH` the
  referenced issue to `state: closed` via the Forgejo API. Do not rely
  on `Closes #N` keyword auto-close — it has not worked reliably in
  this Forgejo instance, leaving issues stale while their PRs are
  merged. Manual close is one extra API call and is part of the merge
  step, not optional.
 - **Push after commits** — keep Forgejo in sync after each commit or logical batch
 ---
@ -104,12 +131,66 @@ only).
 ---
 ## Documentation Workflow
 - **Wiki location:** `docs/wiki/` — local git checkout of `luminos.wiki.git`
 - **Clone URL:** `ssh://git@forgejo-claude/archeious/luminos.wiki.git`
 - **Session startup:** clone if missing, `git -C docs/wiki pull` if present
 - **All reads and writes** happen on local files in `docs/wiki/`. Use Read,
  Edit, Write, Grep, Glob — never the Forgejo web API for wiki content.
 - **Naming:** CamelCase slugs (`Architecture.md`, `DevelopmentGuide.md`).
  Display name comes from the H1 heading inside the file.
 - **Commits:** direct to main branch. Batch logically — commit when finishing
  a round of updates, not after every file.
 - **Push:** after each commit batch.
 ---
 ## ADHD Session Protocols
 > **MANDATORY — follow literally, every session, no exceptions.**
 1. **Session Start Ritual** — Ensure `docs/wiki/` is cloned and current.
   Fetch open issues from Forgejo (`archeious/luminos`) and present them as
   suggested tasks. Ask: *"What's the one thing we're shipping?"* Once the
   user answers, match to an existing issue or create one before starting
   work. Do NOT summarize project state, recap history, or do any other work
   before asking this question.
 2. **Dopamine-Friendly Task Sizing** — break work into 5–15 minute tasks with
   clear, visible outputs. Each task should have a moment of completion.
 3. **Focus Guard** — classify incoming requests as on-topic / adjacent /
   off-topic. Name it out loud before acting. Adjacent work goes to a new
   issue; off-topic work gets deferred.
 4. **Shiny Object Capture** — when a new idea surfaces mid-session, write it
   to PLAN.md (Raw Thoughts) or open a Forgejo issue, then return to the
   current task. Do not context-switch.
 5. **Breadcrumb Protocol** — after each completed task, output:
   `Done: <what was completed>. Next: <what comes next>.`
   This re-orients after any interruption.
 6. **Session End Protocol** — before closing, state the exact pickup point for
   the next session: branch name, file, what was in progress, and the
   recommended first action next time.
 ---
 ## Session Protocols
 - **"externalize"** → read and follow `docs/externalize.md`
 - **"wrap up" / "end session"** → read and follow `docs/wrap-up.md`
 ---
 ## Session Log
 | # | Date | Summary |
 |---|---|---|
-| 6 | 2026-04-07 | Extracted shared workflow/branching/protocols from project CLAUDE.md to global `~/.claude/CLAUDE.md`; moved externalize.md and wrap-up.md to `~/.claude/protocols/` |
+| 2 | 2026-04-06 | Forgejo milestones (9), issues (36), project board, Gitea MCP installed and configured globally |
-| 7 | 2026-04-07 | Phase 1 audit — closed #1 (only #54 remains); gitea MCP credential overhaul: dedicated `claude-code` Forgejo user, admin on luminos, write+delete verified end-to-end |
+| 3 | 2026-04-06 | Phase 1 complete (#1–#3), MCP backend architecture design (Part 10, Phase 3.5), issues #38–#40 opened |
-| 8 | 2026-04-07 | Closed #54 — added confidence/confidence_reason to write_cache tool schema description; Phase 1 milestone now 4/4 complete |
+| 4 | 2026-04-06 | Phase 2 + 2.5 complete (#4–#7, #42, #44), filetype classifier rebuild, context budget metric fix, 8 PRs merged, issues #46/#48/#49/#51 opened |
 Full log: wiki — [Session Retrospectives](https://forgejo.labbity.unbiasedgeek.com/archeious/luminos/wiki/SessionRetrospectives)
--- a/202
+++ b/202
@ -1,202 +0,0 @@
                                 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 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 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 those 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
   APPENDIX: How to apply the Apache License to your work.
      To apply the Apache License to your work, attach the following
      boilerplate notice, with the fields enclosed by brackets "[]"
      replaced with your own identifying information. (Don't include
      the brackets!)  The text should be enclosed in the appropriate
      comment syntax for the file format. We also recommend that a
      file or class name and description of purpose be included on the
      same "printed page" as the copyright notice for easier
      identification within third-party archives.
   Copyright [yyyy] [name of copyright owner]
   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.
--- a/PLAN.md
+++ b/PLAN.md
@ -576,39 +576,12 @@ without 9 phases of rework.
  output size, redundant reads) before picking a fix.
 - Add token-usage instrumentation so regressions are visible.
 ### Phase 2.6 — Pre-Phase-3 cleanup (#54, #57)
 Two debts surfaced during the Session 5 documentation deep dive that
 should be paid before Phase 3 adds more state to the same code paths:
 - **#54 — Phase 1 confidence-write path is dormant.** The cache schema
  accepts `confidence` and `low_confidence_entries()` works, but no
  prompt instructs the agent to set the field. Phase 1 is half-shipped.
  Wire the prompt before Phase 8 (refinement) tries to consume the
  signal — preferably now while Phase 1 context is still fresh.
 - **#57 — Refactor `_run_dir_loop` before #10 lands.** The function is
  at its readability ceiling (~160 lines, four conceptual layers).
  Phase 3 #10 (dynamic turn allocation) will inject per-dir `max_turns`
  and possibly mid-loop renegotiation. Extract `_build_dir_loop_context`,
  `_check_budget_and_flush_partial`, and `_handle_turn_response` into
  focused helpers first; refactoring after #10 will be harder.
 ### Phase 3 — Investigation planning
 - Planning pass after survey, before dir loops
 - `submit_plan` tool
 - Dynamic turn allocation based on plan
 - Dir loop orchestrator updated to follow plan
 ### Phase 3.4 — Tool registration cleanup (#56)
 Adding a tool currently requires updating `_TOOL_DISPATCH` and
 `_DIR_TOOLS` in two separate places in `ai.py`. Forgetting one half is
 silent. A small `@dir_tool` decorator collapses this into one
 declaration. **Decision point:** Phase 3.5 (MCP) will replace
 `_TOOL_DISPATCH` with dynamic discovery from an MCP server, which makes
 this issue partially moot. Either fix now (easier to migrate one
 well-structured registry to MCP than two parallel structures) or defer
 and let it die with the Python dispatch table. Decide before starting
 3.5.
 ### Phase 3.5 — MCP backend abstraction (pivot point)
 See Part 10 for full design. This phase happens *after* Phase 3 is
 working and *before* Phase 4. The goal is to migrate the filesystem
@ -671,14 +644,6 @@ architecture. The migration pain is intentional and instructive.
 - Report formatter renders populated fields only
 - Domain-appropriate section headers
 ### Background chore — Unit tests for pure helpers in ai.py (#55)
 `ai.py` is documented as exempt from unit testing because the dir loop
 needs a live API. But several pure helpers (`_filter_dir_tools`,
 `_format_survey_block`, `_format_survey_signals`, `_default_survey`,
 `_should_skip_dir`, `_path_is_safe`, `_block_to_dict`) have no API
 dependency and can be tested directly. Low priority, not phase-blocking,
 fold into any session that touches these helpers.
 ### End-of-project tuning
 - **Honest terminal report file-type view (#49)** — the report still
  shows the bucketed `summarize_categories()` view, which collapses
--- a/README.md
+++ b/README.md
@ -1,133 +0,0 @@
 # Luminos
 A file system intelligence tool. Scans a directory and produces a reconnaissance report that tells you what the directory is, what's in it, and what might be worth your attention.
 Luminos has two modes. The **base mode** is a single Python file that uses only the standard library and GNU coreutils. No pip install, no virtual environment, no dependencies to audit. The **`--ai` mode** runs a multi-pass agentic investigation against the [Claude API](https://www.anthropic.com/api) to reason about what the project actually does and flag anything that looks off. AI mode is opt-in and is the only path that requires pip-installable packages.
 ## Why
 Most "repo explorer" tools answer one question: "what files are here?" Luminos is built around a harder question: "what is this, and should I be worried about any of it?"
 The base scan gives you the mechanical answer: directory tree, file classification across seven categories, language breakdown with line counts, recently modified files, disk usage, and the largest files. That is usually enough for a quick "what is this" look.
 The AI mode goes further. It runs an isolated investigation per directory, leaves-first, with a small toolbelt (read files, run whitelisted coreutils commands, write cache entries) and a per-directory context budget. Each directory gets its own summary, and a final synthesis pass reads only the directory-level cache entries to produce a whole-project verdict. Findings are flagged with a severity level (`critical`, `concern`, or `info`) so the important stuff floats to the top.
 ## Features
 - **Zero dependencies in base mode.** Runs on bare Python 3 plus the GNU coreutils you already have.
 - **Graceful degradation everywhere.** Permission denied, subprocess timeouts, missing API key, missing optional packages: all handled without crashing the scan.
 - **Directory tree.** Visual tree with configurable depth and exclude patterns.
 - **File classification.** Files bucketed into seven categories (code, config, docs, data, media, binary, other) via `file(1)` magic.
 - **Language detection and LOC counting.** Which languages are present, how many lines of code per language.
 - **Recently modified files.** Surface the files most likely to reflect current activity.
 - **Disk usage.** Per-directory disk usage with top offenders called out.
 - **Watch mode.** Re-scan every 30 seconds and show diffs.
 - **JSON output.** Pipe reports to other tools or save for comparison.
 - **AI investigation (opt-in).** Multi-pass, leaves-first agentic analysis via Claude, with an investigation cache so repeat runs are cheap.
 - **Severity-ranked flags.** Findings are sorted so `critical` items are the first thing you see.
 ## Installation
 ### Base mode
 No installation required. Clone and run.
 ```bash
 git clone https://github.com/archeious/luminos.git
 cd luminos
 python3 luminos.py <target>
 ```
 Works on any system with Python 3 and standard GNU coreutils (`wc`, `file`, `grep`, `head`, `tail`, `stat`, `du`, `find`).
 ### AI mode
 AI mode needs a few pip-installable packages. The project ships a helper script that creates a dedicated virtual environment and installs them:
 ```bash
 ./setup_env.sh
 source ~/luminos-env/bin/activate
 ```
 The packages installed are `anthropic`, `tree-sitter`, a handful of tree-sitter language grammars, and `python-magic`.
 You also need an Anthropic API key exported as an environment variable:
 ```bash
 export ANTHROPIC_API_KEY=your-key-here
 ```
 Check which optional dependencies are present:
 ```bash
 python3 luminos.py --install-extras
 ```
 ## Usage
 ### Base scan
 ```bash
 python3 luminos.py /path/to/project
 ```
 ### AI scan
 ```bash
 python3 luminos.py --ai /path/to/project
 ```
 ### Common flags
 ```bash
 # Deeper tree, include hidden files, exclude build and vendor dirs
 python3 luminos.py -d 8 -a -x .git -x node_modules -x vendor /path/to/project
 # JSON output to a file
 python3 luminos.py --json -o report.json /path/to/project
 # Watch mode (re-scan every 30s, show diffs)
 python3 luminos.py --watch /path/to/project
 # Force a fresh AI investigation, ignoring the cache
 python3 luminos.py --ai --fresh /path/to/project
 # Clear the AI investigation cache
 python3 luminos.py --clear-cache
 ```
 Run `python3 luminos.py --help` for the full flag list.
 ## How AI mode works
 A short version of what happens when you pass `--ai`:
 1. **Discover** every directory under the target.
 2. **Sort leaves-first** so the deepest directories are investigated before their parents.
 3. **Run an isolated agent loop per directory** with a max of 10 turns each. The agent has a small toolbelt: read files, run whitelisted coreutils commands (`wc`, `file`, `grep`, `head`, `tail`, `stat`, `du`, `find`), and write cache entries.
 4. **Cache everything.** Each file and directory summary is written to `/tmp/luminos/` so that subsequent runs on the same target don't burn tokens re-deriving things that haven't changed.
 5. **Context budget guard.** Per-turn `input_tokens` is watched against a budget (currently 70% of the model's context window) so a rogue directory can't blow the context and silently degrade quality.
 6. **Final synthesis pass** reads only the directory-level cache entries (not the raw files) to produce a project-level summary and the severity-ranked flags.
 ## Development
 Run the test suite:
 ```bash
 python3 -m unittest discover -s tests/
 ```
 Modules that are intentionally not unit tested and why:
 - `luminos_lib/ai.py`: requires a live Anthropic API, tested in practice
 - `luminos_lib/ast_parser.py`: requires tree-sitter grammars installed
 - `luminos_lib/watch.py`: stateful event loop, tested manually
 - `luminos_lib/prompts.py`: string templates only
 ## License
 Apache License 2.0. See [`LICENSE`](LICENSE) for the full text.
 ## Source of truth
 The canonical home for this project is the [Forgejo repository](https://forgejo.labbity.unbiasedgeek.com/archeious/luminos). The GitHub copy is a read-only mirror, pushed automatically from Forgejo. Issues, pull requests, and the project wiki live on Forgejo.
--- a/docs/externalize.md
+++ b/docs/externalize.md
@ -0,0 +1,36 @@
 # Externalize Protocol
 > Triggered when the user says "externalize" or "externalize your thoughts."
 > This is a STANDALONE action. Do NOT wrap up unless separately asked.
 ## Steps
 1. **Determine session number** — check the Session Log in CLAUDE.md for the
   latest session number, increment by 1
 2. **Pull wiki** — ensure `docs/wiki/` is current:
   ```bash
   git -C docs/wiki pull   # or clone if missing
   ```
 3. **Create session wiki page** — write `docs/wiki/Session{N}.md` with:
   - Date, focus, duration estimate
   - What was done (with detail — reference actual files and commits)
   - Discoveries and observations
   - Decisions made and why
   - Raw Thinking — observations, concerns, trade-offs, and loose threads that
     came up during the session but weren't part of the main deliverable.
     Things you'd mention if pair programming: prerequisites noticed, corners
     being painted into, intent mismatches, unresolved questions.
   - What's next
 4. **Update SessionRetrospectives.md** — read the current index, add the new
   session row, write it back
 5. **Commit and push wiki:**
   ```bash
   cd docs/wiki
   git add -A
   git commit -m "retro: Session {N} — <one-line summary>"
   git push
   ```
--- a/docs/wrap-up.md
+++ b/docs/wrap-up.md
@ -0,0 +1,31 @@
 # Session Wrap-Up Checklist
 > Triggered when the user says "wrap up", "end session", or similar.
 > Always externalize FIRST, then do the steps below.
 ## Steps
 1. **Externalize** — run the `docs/externalize.md` protocol if not already
   done this session
 2. **Reread CLAUDE.md** — ensure you have the latest context before editing
 3. **Update CLAUDE.md:**
   - Update **Current Project State** — phase, last worked on (today's date),
     last commit, blocking issues
   - Update **Session Log** — add new entry, keep only last 3 sessions,
     remove older ones (full history is in the wiki)
 4. **Commit and push main repo:**
   ```bash
   git add CLAUDE.md
   git commit -m "chore: update CLAUDE.md for session {N}"
   git push
   ```
 5. **Verify nothing is unpushed** — both the main repo and docs/wiki should
   have no pending commits
 6. **Recommend next session** — tell the user what the best next session
   should tackle, in priority order based on PLAN.md and any open Forgejo
   issues
--- a/luminos_lib/ai.py
+++ b/luminos_lib/ai.py
@ -247,14 +247,9 @@ _DIR_TOOLS = [
                    "type": "object",
                    "description": (
                        "Cache entry. Files: {path, relative_path, size_bytes, "
-                        "category, summary, notable, notable_reason, "
+                        "category, summary, notable, notable_reason, cached_at}. "
                        "confidence, confidence_reason, cached_at}. "
                        "Dirs: {path, relative_path, child_count, summary, "
-                        "dominant_category, notable_files, "
+                        "dominant_category, notable_files, cached_at}."
                        "confidence, confidence_reason, cached_at}. "
                        "Always set confidence (0.0–1.0); see system prompt "
                        "for calibration. Set confidence_reason only when "
                        "confidence < 0.7."
                    ),
                },
            },