Separate live Senpai skills from operator guides

.agents/agents/search.md

@@ -10,9 +10,6 @@ permission_mode: never_confirm
 tools:
   - terminal
   - file_editor
-skills:
-  - exa-search
-  - alphaxiv-paper-lookup
 ---
 
 You are Senpai's external research agent. The delegated prompt begins with one

.agents/skills/analyze-experiments/SKILL.md

@@ -1,7 +1,7 @@
 ---
 name: analyze-experiments
 description: >
-  Analyzes and categorizes all ML experiment PRs in the senpai research track.
+  Operator-side analysis of historical ML experiment PRs in Senpai research tracks.
   Use this skill whenever the user asks to: analyze experiments, categorize PRs, bucket
   experiments, summarize what's been tried, understand experiment history, review merged
   vs closed results, or asks "what experiments have we run / worked / failed".
@@ -13,9 +13,11 @@ description: >
 
 # Analyze Experiments Skill
 
-Fetches fresh experiment PR data via the list-experiments skill, categorizes each PR
-using a team of parallel readers, and produces a 5-section report: full catalogue,
-category breakdown with merge rates, merged-only wins, closed-only failures, key narratives.
+This is a human/operator analysis guide, not a live advisor or student skill.
+It fetches fresh experiment PR data via the list-experiments skill, categorizes
+each PR using parallel readers, and produces a 5-section report: full catalogue,
+category breakdown with merge rates, merged-only wins, closed-only failures,
+and key narratives.
 
 ---
 

.agents/skills/git-research-log/SKILL.md

@@ -1,15 +1,16 @@
 ---
 name: git-research-log
 description: >
-  How to document and publish ML experiment results to GitHub as a pull request.
-  Use this skill whenever a student agent has finished running experiments and
-  needs to create or update a PR — even if the instruction is just "wrap up" or
-  "log your results" or "open a PR". Also use it mid-session to update an
-  in-progress PR as trials complete.
+  Developer reference for Senpai's retired direct-GitHub experiment logging
+  workflow. Use only when auditing or migrating legacy research tracks.
 ---
 
 # Experiment PR Skill
 
+This guide is not installed into live advisor or student runtimes. Current
+students use the plugin's `submit-experiment-results` skill and typed result
+tool instead.
+
 The master agent creates an empty PR and hands it to you. Your job is to fill it out as you run trials and finalize it when you're done. The master reads these PRs to decide what to explore next — completeness and honest analysis matter more than polish.
 
 ## PR Title Format

.agents/skills/list-experiments/SKILL.md

@@ -1,11 +1,13 @@
 ---
 name: list-experiments
-description: Use this skill whenever you need to list all of the experiment ideas tried and in progress for this research programme. It outputs 3 files organized by usefulness — merged winners, a compact results table, and full details for deep dives. Use when generating new experimental ideas to check what has already been tried.
+description: Operator-side inventory of experiment PRs for auditing a Senpai research track. It outputs merged winners, a compact results table, and full details for deep dives.
 ---
 
 # List Experiments
 
-Run the script below to fetch all experiment PRs from the advisor branch and organize them into 3 files.
+This is a human/operator audit guide, not a live advisor or student skill. Run
+the script below to fetch all experiment PRs from the advisor branch and
+organize them into 3 files.
 
 The `BASE_BRANCH` should be set to the advisor branch (e.g. `noam`). Check the `$ADVISOR_BRANCH` env var or the PR base branch.
 

.agents/skills/plot-experiment-charts/SKILL.md

@@ -5,27 +5,28 @@
 
 name: plot-experiment-charts
 description: >
-  Generate a training curve comparison chart and embed it in a GitHub PR description.
-  Use this skill whenever a student has finished running experiments and is preparing
-  to submit their PR for advisor review. Triggers on: "plot training curves", "add chart
-  to PR", "visualize experiment", "training curve comparison", "plot-experiment-charts",
-  "add chart", "generate comparison chart". Run this before marking the PR ready for review.
+  Operator-side guide for generating a target-specific training curve chart
+  for a legacy experiment PR. Use when auditing or presenting historical runs.
 ---
 
 # Plot Experiment Charts
 
-You've just finished one or more training runs. Before submitting for review, generate a
-comparison chart so the advisor can see the training dynamics at a glance — not just the
-final numbers, but how the experiment got there. A bolded best-run line and a properly
-scaled y-axis make the story immediately readable, even if some runs diverged.
+This guide is not installed into live advisor or student runtimes. Target
+repositories that require experiment charts should provide their own project
+skill with the correct metrics and plotting code.
+
+Generate a comparison chart so a reviewer can see the training dynamics at a
+glance—not just the final numbers, but how the experiment got there. A bolded
+best-run line and a properly scaled y-axis make the story immediately readable,
+even if some runs diverged.
 
 This skill takes about 30 seconds. It's worth it.
 
 ## What you need
 
 - **Baseline W&B run ID**: in the PR body under `## Baseline`, look for the `W&B run: \`xxxxxxxx\`` line.
 - **Your own run IDs**: the 8-character W&B IDs of the runs you just completed. Find them in the W&B run URLs or in the training output (the run ID is printed at launch).
-- **W&B credentials**: `WANDB_ENTITY` and `WANDB_PROJECT` env vars (already set in the pod environment).
+- **W&B credentials**: `WANDB_ENTITY` and `WANDB_PROJECT` in the current environment.
 
 ## Step 1 — Run the script
 

AGENTS.md

@@ -19,7 +19,7 @@ When helping a user onboard a target repository, inspect an explicitly configure
 - operational guardrails such as commands, budgets, allowed edits, protected artifacts, and result-reporting expectations; and
 - optional research avenues, papers, models, and libraries that provide useful starting points without forcing a narrow solution path.
 
-Favor high-level goals and guardrails that let research agents discover the details. Avoid micromanaging methods or over-prompting one idea unless that narrow focus is the user's explicit goal. The [`bootstrap-target`](plugins/senpai/skills/bootstrap-target/SKILL.md) guide and its template can turn the confirmed decisions into the target contract.
+Favor high-level goals and guardrails that let research agents discover the details. Avoid micromanaging methods or over-prompting one idea unless that narrow focus is the user's explicit goal. The [`bootstrap-target`](.agents/skills/bootstrap-target/SKILL.md) guide and its template can turn the confirmed decisions into the target contract.
 
 Reference examples:
 

README.md

@@ -303,27 +303,30 @@ Children share the parent workspace, so their process and conversation are isola
 
 ## Task guides
 
-OpenHands receives these as progressively disclosed skills; their bodies are loaded only when the task calls for them.
+Live advisors and students load Senpai-owned skills only from
+`plugins/senpai/skills`; target repositories may also supply project skills.
+Guides under `.agents/skills` are for human users and Senpai developers and
+are not installed into autoresearch pods.
 
-### Core research workflow
+### Live autoresearch
 
 | Guide | Purpose |
 |---|---|
-| [Bootstrap a target](plugins/senpai/skills/bootstrap-target/SKILL.md) | Build `program.md` from a new ML repository. |
 | [Assign an experiment](plugins/senpai/skills/assign-experiment/SKILL.md) | Turn a hypothesis into a typed student branch and draft PR. |
 | [Delegate subagents](plugins/senpai/skills/delegate-subagents/SKILL.md) | Launch and coordinate bounded parallel research, review, and implementation help. |
 | [Submit experiment results](plugins/senpai/skills/submit-experiment-results/SKILL.md) | Commit the tested implementation and publish a structured, evidence-backed result. |
 | [Review an experiment](plugins/senpai/skills/review-experiment/SKILL.md) | Merge a reproducible winner, close a useful negative, or request the missing evidence. |
 | [Handle human Issues](plugins/senpai/skills/check-human-issues/SKILL.md) | Respond to authenticated human-to-agent messages delivered through GitHub Issues. |
+| [Senpai status check](plugins/senpai/skills/senpai-status-check/SKILL.md) | Produce a bounded, read-only GitHub, W&B, and local-controller status report. |
+| [Exa search](plugins/senpai/skills/exa-search/SKILL.md) | Search the current web or scholarly publications with mode-specific defaults. |
+| [AlphaXiv paper lookup](plugins/senpai/skills/alphaxiv-paper-lookup/SKILL.md) | Get a structured overview before reading a primary paper deeply. |
+| [W&B and Weave](plugins/senpai/skills/wandb-primary/SKILL.md) | Inspect runs, metrics, artifacts, evaluations, and agent traces. |
 
-### Evidence and research
+### Human and developer guides
 
 | Guide | Purpose |
 |---|---|
-| [Senpai status check](.agents/skills/senpai-status-check/SKILL.md) | Produce a bounded, read-only GitHub, W&B, and local-controller status report. |
-| [Exa search](.agents/skills/exa-search/SKILL.md) | Search the current web or scholarly publications with mode-specific defaults. |
-| [AlphaXiv paper lookup](.agents/skills/alphaxiv-paper-lookup/SKILL.md) | Get a structured overview before reading a primary paper deeply. |
-| [W&B and Weave](.agents/skills/wandb-primary/SKILL.md) | Inspect runs, metrics, artifacts, evaluations, and agent traces. |
+| [Bootstrap a target](.agents/skills/bootstrap-target/SKILL.md) | Build `program.md` from a new ML repository. |
 | [Experiment report](.agents/skills/experiment-report/SKILL.md) | Create the project-standard `nn_cfd` W&B comparison report; this guide is target-specific rather than part of the generic runtime. |
 | [Training code style](literature_and_guidance/TRAINING-CODE-STYLE.md) | Structure expensive ML entrypoints so configuration, artifacts, validation, and failure boundaries stay explicit. |
 

SPEC.md

@@ -215,7 +215,10 @@ suffix on every inference, and current time is rendered for every controller
 wake. Operators must start fresh role state to apply a changed identity,
 program, or role charter.
 
-File-based subagents are discovered from `.agents/agents`. Skill bodies are not
+File-based subagents are discovered from `.agents/agents`. Live advisor and
+student skills come only from `plugins/senpai/skills`; `.agents/skills` is for
+human operators and Senpai developers and is not installed into pods. Target
+repositories may still supply their own project skills. Skill bodies are not
 concatenated into agent definitions. The OpenHands fork's `main` branch applies
 each agent definition's `reasoning_effort` override after resolving its
 inherited LLM or stored model profile.
@@ -687,7 +690,8 @@ Removed:
 
 Retained intentionally:
 
-- Agent skills and their model/effort metadata under `.agents`;
+- runtime skills and their model/effort metadata in the Senpai plugin;
+- human and developer guides under `.agents/skills`, outside pod context;
 - OpenHands Browser, task tracker, Think, and the high-quality default
   condenser for providers not using stored OpenAI Responses continuation or
   Anthropic native compaction;

k8s/entrypoint-advisor.sh

@@ -129,7 +129,7 @@ ls \
     "$HOME/.agents/agents/general-purpose.md" \
     "$HOME/.agents/agents/explore.md" \
     "$HOME/.agents/agents/search.md" \
-    "$HOME/.agents/skills/wandb-primary/SKILL.md"
+    "$SENPAI_PLUGIN/skills/wandb-primary/SKILL.md"
 
 # --- Hivemind is intentionally disabled pending its OpenHands rewrite. ---
 # source "$WORKDIR/k8s/start-hivemind.sh"

k8s/entrypoint-student.sh

@@ -79,7 +79,7 @@ ls \
     "$HOME/.agents/agents/general-purpose.md" \
     "$HOME/.agents/agents/explore.md" \
     "$HOME/.agents/agents/search.md" \
-    "$HOME/.agents/skills/wandb-primary/SKILL.md"
+    "$SENPAI_PLUGIN/skills/wandb-primary/SKILL.md"
 
 # --- Hivemind is intentionally disabled pending its OpenHands rewrite. ---
 # source "$WORKDIR/k8s/start-hivemind.sh"

plugins/senpai/README.md

@@ -32,7 +32,10 @@ no MCP server. The Python runtime binds the authenticated role and adds the
 canonical `ADVISOR:` or `STUDENT:` prefix to Senpai-authored GitHub comments;
 tool payloads contain only the unprefixed message text.
 
-Keep Senpai-owned workflow skills here rather than relying on a provider's user
-skill directory. Never commit secret values. The plugin remains the source of
-truth for reusable workflow guidance, while Python remains the source of truth
-for verified state changes.
+Keep every Senpai-owned skill used by a live advisor or student here rather
+than relying on a provider's user skill directory. Target repositories may
+supply project skills separately. Human onboarding and developer guides stay
+under the runner's `.agents/skills` and are not installed into pods. Never
+commit secret values. The plugin remains the source of truth for reusable
+runtime guidance, while Python remains the source of truth for verified state
+changes.

plugins/senpai/scripts/agent-context.sh

@@ -10,15 +10,10 @@ install_senpai_agent_context() {
     local runtime_root="$3"
     local runtime_plugin="$runtime_root/plugin"
 
-    mkdir -p "$HOME/.agents/skills"
-    cp -a "$workdir/.agents/." "$HOME/.agents/"
-    for marker in "$HOME"/.agents/skills/*/.senpai-developer-only; do
-        [ -e "$marker" ] || continue
-        rm -rf -- "${marker%/.senpai-developer-only}"
-    done
+    mkdir -p "$HOME/.agents/agents"
+    cp -a "$workdir/.agents/agents/." "$HOME/.agents/agents/"
     cp -a "$source_plugin" "$runtime_plugin"
     "$SENPAI_PYTHON" -m senpai_agent.agent_markdown \
-        "$HOME/.agents" "$runtime_plugin"
-    cp -a "$runtime_plugin/skills/." "$HOME/.agents/skills/"
+        "$HOME/.agents/agents" "$runtime_plugin"
     printf '%s\n' "$runtime_plugin"
 }

.agents/skills/exa-search/SKILL.md → plugins/senpai/skills/exa-search/SKILL.md

@@ -10,13 +10,13 @@ official `exa_py` client, loads the nearest `.env` through `python-dotenv`, and
 preserves an `EXA_API_KEY` already set in the environment.
 
 ```bash
-python "$HOME/.agents/skills/exa-search/scripts/search_exa.py" \
+python "$SENPAI_PLUGIN/skills/exa-search/scripts/search_exa.py" \
   general-web \
   "current OpenHands SDK file-based agent documentation"
 ```
 
 ```bash
-python "$HOME/.agents/skills/exa-search/scripts/search_exa.py" \
+python "$SENPAI_PLUGIN/skills/exa-search/scripts/search_exa.py" \
   research-publications \
   "uncertainty calibration for neural networks"
 ```

.agents/skills/wandb-primary/SKILL.md → plugins/senpai/skills/wandb-primary/SKILL.md

@@ -59,13 +59,13 @@ This skill covers everything an agent needs to work with Weights & Biases:
 
 | I need to... | Use |
 |---|---|
-| Query training runs, loss curves, hyperparameters | **W&B SDK** (`wandb.Api()`) — see `references/WANDB_SDK.md` |
-| Query GenAI traces, calls, evaluations | **Weave SDK** (`weave.init()`, `client.get_calls()`) — see `references/WEAVE_SDK.md` |
+| Query training runs, loss curves, hyperparameters | **W&B SDK** (`wandb.Api()`) — see `$SENPAI_PLUGIN/skills/wandb-primary/references/WANDB_SDK.md` |
+| Query GenAI traces, calls, evaluations | **Weave SDK** (`weave.init()`, `client.get_calls()`) — see `$SENPAI_PLUGIN/skills/wandb-primary/references/WEAVE_SDK.md` |
 | Convert Weave wrapper types to plain Python | **`weave_helpers.unwrap()`** |
 | Build a DataFrame from training runs | **`wandb_helpers.runs_to_dataframe()`** |
 | Extract eval results for analysis | **`weave_helpers.eval_results_to_dicts()`** |
-| Need low-level Weave filtering (CallsFilter, Query) | **Raw Weave SDK** (`weave.init()`, `client.get_calls()`) — see `references/WEAVE_SDK.md` |
-| Judge curve shape (spikes, smoothness, slope, overfit) | **`training_diagnostics` + `curve_plots`** — use the workflow below, then load `references/TRAINING_DIAGNOSTICS.md` for the heuristics |
+| Need low-level Weave filtering (CallsFilter, Query) | **Raw Weave SDK** (`weave.init()`, `client.get_calls()`) — see `$SENPAI_PLUGIN/skills/wandb-primary/references/WEAVE_SDK.md` |
+| Judge curve shape (spikes, smoothness, slope, overfit) | **`training_diagnostics` + `curve_plots`** — use the workflow below, then load `$SENPAI_PLUGIN/skills/wandb-primary/references/TRAINING_DIAGNOSTICS.md` for the heuristics |
 
 ---
 
@@ -74,8 +74,9 @@ This skill covers everything an agent needs to work with Weights & Biases:
 ### Helper libraries
 
 ```python
+import os
 import sys
-sys.path.insert(0, ".agents/skills/wandb-primary/scripts")
+sys.path.insert(0, f"{os.environ['SENPAI_PLUGIN']}/skills/wandb-primary/scripts")
 
 # Weave helpers (traces, evals, GenAI)
 from weave_helpers import (
@@ -125,9 +126,9 @@ from curve_plots import (
 
 Read these as needed — they contain full API surfaces and recipes:
 
-- **`references/WEAVE_SDK.md`** — Weave SDK for GenAI traces (`client.get_calls()`, `CallsFilter`, `Query`, stats). Start here for Weave queries.
-- **`references/WANDB_SDK.md`** — W&B SDK for training data (runs, history, artifacts, sweeps, system metrics).
-- **`references/TRAINING_DIAGNOSTICS.md`** — reference heuristics for reading loss / LR / grad-norm / grad-histogram charts. Load this when you are actively interpreting training curves.
+- **`$SENPAI_PLUGIN/skills/wandb-primary/references/WEAVE_SDK.md`** — Weave SDK for GenAI traces (`client.get_calls()`, `CallsFilter`, `Query`, stats). Start here for Weave queries.
+- **`$SENPAI_PLUGIN/skills/wandb-primary/references/WANDB_SDK.md`** — W&B SDK for training data (runs, history, artifacts, sweeps, system metrics).
+- **`$SENPAI_PLUGIN/skills/wandb-primary/references/TRAINING_DIAGNOSTICS.md`** — reference heuristics for reading loss / LR / grad-norm / grad-histogram charts. Load this when you are actively interpreting training curves.
 
 ---
 
@@ -216,7 +217,7 @@ df = pd.DataFrame(rows)
 print(df.describe())
 ```
 
-For full W&B SDK reference (filters, history, artifacts, sweeps), read `references/WANDB_SDK.md`.
+For full W&B SDK reference (filters, history, artifacts, sweeps), read `$SENPAI_PLUGIN/skills/wandb-primary/references/WANDB_SDK.md`.
 
 ### Weave — SDK
 
@@ -226,7 +227,7 @@ client = weave.init(f"{entity}/{project}")  # positional string, NOT keyword arg
 calls = client.get_calls(limit=10)
 ```
 
-For raw SDK patterns (CallsFilter, Query, advanced filtering), read `references/WEAVE_SDK.md`.
+For raw SDK patterns (CallsFilter, Query, advanced filtering), read `$SENPAI_PLUGIN/skills/wandb-primary/references/WEAVE_SDK.md`.
 
 ---
 
@@ -307,7 +308,7 @@ For structured failure analysis on eval results:
 3. **Axial coding** — write a second Scorer that classifies notes into a taxonomy
 4. **Summarize** — count primary labels with `collections.Counter`
 
-See `references/WEAVE_SDK.md` for the full SDK reference.
+See `$SENPAI_PLUGIN/skills/wandb-primary/references/WEAVE_SDK.md` for the full SDK reference.
 
 ### W&B Reports
 
@@ -344,7 +345,7 @@ Keep the inline workflow short and load detail on demand:
 1. Confirm `step_key` before doing any curve work. Never assume `_step`.
 2. Compute features with the bundled helpers instead of hand-rolling spike or slope logic.
 3. Render PNGs and inspect them visually.
-4. Load `references/TRAINING_DIAGNOSTICS.md` while you interpret the results.
+4. Load `$SENPAI_PLUGIN/skills/wandb-primary/references/TRAINING_DIAGNOSTICS.md` while you interpret the results.
 5. End with a verdict, evidence tied to step ranges, and concrete next actions.
 
 ### Required sequence
@@ -370,7 +371,7 @@ Next actions:
 - <concrete hyperparameter, logging, or code change>
 ```
 
-Load `references/TRAINING_DIAGNOSTICS.md` for the interpretation heuristics, especially when the numbers and the image disagree.
+Load `$SENPAI_PLUGIN/skills/wandb-primary/references/TRAINING_DIAGNOSTICS.md` for the interpretation heuristics, especially when the numbers and the image disagree.
 
 ---
 

.agents/skills/wandb-primary/scripts/wandb_helpers.py → plugins/senpai/skills/wandb-primary/scripts/wandb_helpers.py

@@ -9,8 +9,9 @@
 run data into pandas-friendly structures for analysis.
 
 Usage (in sandbox):
+    import os
     import sys
-    sys.path.insert(0, ".agents/skills/wandb-primary/scripts")
+    sys.path.insert(0, f"{os.environ['SENPAI_PLUGIN']}/skills/wandb-primary/scripts")
     from wandb_helpers import (
         runs_to_dataframe,   # Convert runs to a clean pandas DataFrame
         diagnose_run,        # Quick diagnostic summary of a training run