@monoes/monomindcli 1.10.47 → 1.10.55

package/.claude/skills/mastermind/approve.md

@@ -97,7 +97,7 @@ approval=$(jq --arg id "$approval_id" '.approvals[] | select(.id == $id)' "$appr
 agent_id=$(echo "$approval" | jq -r '.agent_id')
 memNs="org:${org_name}"
 npx monomind@latest memory store \
-  --key "${memNs}:approval:${approval_id}" \
+  --key "approval:${approval_id}" \
   --namespace "$memNs" \
   --value '{"status":"approved","approval_id":"'"$approval_id"'","ts":"'"$(date -u +%Y-%m-%dT%H:%M:%SZ)"'"}'
 
@@ -118,7 +118,7 @@ approval=$(jq --arg id "$approval_id" '.approvals[] | select(.id == $id)' "$appr
 agent_id=$(echo "$approval" | jq -r '.agent_id')
 memNs="org:${org_name}"
 npx monomind@latest memory store \
-  --key "${memNs}:approval:${approval_id}" \
+  --key "approval:${approval_id}" \
   --namespace "$memNs" \
   --value '{"status":"rejected","approval_id":"'"$approval_id"'","reason":"'"${reason:-No reason given}"'","ts":"'"$(date -u +%Y-%m-%dT%H:%M:%SZ)"'"}'
 
@@ -150,13 +150,21 @@ jq --arg id "$approval_id" \
 
 2. Emit `org:approval:requested` event to dashboard (so the human is notified).
 
-3. Poll memory for the approval decision:
+3. Poll for the approval decision — check the approvals file first (authoritative, updated by both the dashboard UI and `/mastermind:approve`), then fall back to memory:
 ```bash
+approvalsFile=".monomind/orgs/${orgName}-approvals.json"
 while true; do
-  result=$(npx monomind@latest memory search --query "approval:${approval_id}" --namespace "${memNs}" 2>/dev/null)
-  status=$(echo "$result" | jq -r '.[0].value.status // ""' 2>/dev/null)
-  [ "$status" = "approved" ] && break
-  [ "$status" = "rejected" ] && { echo "Action rejected: $(echo $result | jq -r '.[0].value.reason // ""')"; exit 1; }
+  # Check file first — dashboard approve/reject button only updates the file (not memory)
+  file_status=$(jq --arg id "$approval_id" '.approvals[] | select(.id == $id) | .status // ""' "$approvalsFile" 2>/dev/null || echo "")
+  if [ -z "$file_status" ] || [ "$file_status" = '"pending"' ] || [ "$file_status" = "pending" ]; then
+    mem_result=$(npx monomind@latest memory search --query "approval:${approval_id}" --namespace "${memNs}" 2>/dev/null)
+    mem_status=$(echo "$mem_result" | jq -r '.[0].value.status // ""' 2>/dev/null)
+    [ -n "$mem_status" ] && [ "$mem_status" != "pending" ] && file_status="$mem_status"
+  fi
+  # Strip quotes from jq output
+  file_status=$(echo "$file_status" | tr -d '"')
+  [ "$file_status" = "approved" ] && break
+  [ "$file_status" = "rejected" ] && { echo "Action rejected by governance policy — skip this action"; exit 1; }
   sleep 30
 done
 ```

package/.claude/skills/mastermind/autodev.md

@@ -0,0 +1,534 @@
+---
+name: mastermind-autodev
+description: Mastermind autodev — autonomous research → build → review loop. Researches the project, picks the best improvement idea, builds it, then reviews with mastermind:review --tillend until clean. Repeats N times when given a count. Supports --newfeature N for full feature pipeline (research → build → review → document → deliver).
+type: domain-skill
+default_mode: auto
+---
+
+# Mastermind Autodev Domain
+
+This skill is invoked directly via `/mastermind:autodev`.
+
+Autodev is a fully autonomous loop: it researches the project, selects the highest-value improvement, builds it, and reviews it until clean — then repeats.
+
+When `--newfeature N` is passed, it switches to **feature mode**: discovers the N best genuinely-new features the project is missing, then for each one runs the complete pipeline — Build → Review → Documentation → Delivery.
+
+---
+
+## Inputs
+
+- `brain_context`: BRAIN CONTEXT block (loaded via _protocol.md Brain Load Procedure)
+- `count`: number of improvements to build (default: 1; set by leading integer in arguments, e.g. `9 --tillend`)
+- `newfeature_count`: when `--newfeature N` is set, number of brand-new features to discover and fully deliver (overrides normal improvement loop)
+- `focus`: optional topic hint (e.g. "performance", "security", "dx") — narrows research phase
+- `mode`: auto | confirm (default: auto)
+- `current_rep` / `loop_id`: injected by _repeat.md on continuation runs
+
+---
+
+## Flag Parsing
+
+Before anything else, extract from `$ARGUMENTS`:
+
+1. **Leading integer** — if the first token is a bare integer (e.g. `9`), set `count = 9`. Remove it from the remaining args.
+2. **`--count <N>`** — alternate way to set count.
+3. **`--newfeature <N>`** — activate feature mode; set `newfeature_count = N`. When present, the normal improvement loop is **replaced** by the Feature Pipeline below. N is how many new features to discover and deliver end-to-end (must be ≥ 1).
+4. **`--focus <topic>`** — set focus topic.
+5. **`--auto` / `--confirm`** — mode.
+6. **`--tillend`, `--repeat`, `--maxruns`, `--wait`, `--rep`, `--loop`** — handled by _repeat.md PREAMBLE before this skill runs.
+7. Remaining text → `focus` hint (if no `--focus` was given).
+
+If `count` is not set, default to `1`.
+If `--newfeature` is present but N is missing or not a positive integer, default `newfeature_count = 3`.
+If `newfeature_count > 10`: emit `[autodev] Warning: --newfeature capped at 10 (requested: N)` and set `newfeature_count = 10` before proceeding.
+If `--newfeature` and `--tillend` are both present: emit `[autodev] Warning: --tillend is not supported with --newfeature and will be ignored.` and strip the tillend flag before proceeding.
+
+---
+
+## Feature Pipeline (--newfeature mode)
+
+When `--newfeature N` is parsed, skip the normal improvement loop entirely and run this pipeline instead.
+
+### FP-0 — Feature Discovery
+
+**Goal:** Find the N best brand-new features the project is missing — not improvements to existing code, not bugfixes, not refactors. Things that would be genuinely new capability.
+
+Run all research in parallel:
+
+```bash
+git log --oneline -30
+find . -maxdepth 3 -type f \( -name "*.ts" -o -name "*.js" -o -name "*.py" -o -name "*.go" -o -name "*.md" \) \
+  ! -path "*/node_modules/*" ! -path "*/.git/*" ! -path "*/dist/*" | head -80
+cat package.json 2>/dev/null || cat pyproject.toml 2>/dev/null || cat go.mod 2>/dev/null || true
+cat README.md 2>/dev/null | head -120 || true
+# Scan for capability gaps / user-facing TODOs
+grep -rn "TODO\|could add\|future\|roadmap\|planned\|wish\|missing\|not yet\|coming soon" \
+  --include="*.ts" --include="*.js" --include="*.py" --include="*.md" \
+  . 2>/dev/null | grep -v node_modules | grep -v dist | head -40
+```
+
+Also call (parallel):
+- `mcp__monomind__monograph_god_nodes` — find the most-connected modules (surfaces where new features plug in best)
+- `mcp__monomind__monograph_suggest` with query `"missing feature new capability gap"` — graph-guided gaps
+- `mcp__monomind__memory_search` with query `"autodev newfeature built"`, namespace `"mastermind:autodev"`, limit 30 — exclude already-delivered features
+
+**Produce a ranked shortlist of up to `newfeature_count` features.** Aim for exactly `newfeature_count`, but stop at fewer only if the project genuinely has no more viable new features. After producing the shortlist — regardless of mode — **set `newfeature_count = len(shortlist)`** so the loop counter and FP-End summary are accurate. For each:
+
+```
+Feature <K>:
+  title: <short imperative name, e.g. "Add batch export command">
+  user_value: <one sentence — what a user gains>
+  entry_point: <where in the codebase it plugs in (file/module)>
+  feasibility: high | medium
+  effort_estimate: small (< 100 LOC) | medium (100-400 LOC) | large (400+ LOC)
+  type: feature
+```
+
+Priority ranking:
+1. High feasibility first
+2. High user_value for the effort
+3. Not in recent git history
+4. Not in autodev memory (already built)
+5. Aligns with `focus` hint if given
+
+If `mode = confirm`: display the full ranked shortlist and ask the user to approve before continuing. The user may strike features, reorder, or add constraints. After the user responds, update `approved_shortlist` to contain only the features the user kept, then **set `newfeature_count = len(approved_shortlist)`** before starting the FP-1..N loop. The FP-End summary denominator uses this updated count.
+
+If `mode = auto`: `approved_shortlist = shortlist` (all features). `newfeature_count` was already aligned to `len(shortlist)` above — it does not revert to the originally parsed value.
+
+Store the shortlist in memory:
+```
+mcp__monomind__memory_store(
+  content: "autodev newfeature shortlist: [<title1>, <title2>, ...]. Session: <date>.",
+  namespace: "mastermind:autodev",
+  tags: ["autodev", "newfeature", "shortlist"]
+)
+```
+
+---
+
+### FP-1 through FP-N — Full Delivery Loop (one per feature)
+
+Initialize the loop counter before starting:
+```bash
+K=0
+```
+
+For **each feature** in the ranked shortlist (1 through `newfeature_count`), run these phases in order. At the start of each iteration, increment K:
+```bash
+K=$((K + 1))
+```
+
+#### Phase A — Build
+
+(`K` is the current loop index integer — e.g. `1`, `2`, `3`. Treat `$K` and `$FEATURE_FILE_LIST` as template values: re-emit them with the concrete value substituted at the top of each phase's bash block, since each shell invocation is independent. E.g. at the top of Phase B's block: `K=1; FEATURE_FILE_LIST="/tmp/autodev_feature_files_${K}.txt"`. The same rule applies to FP-End: assign `K=<final_newfeature_count>` at the top of the FP-End bash block before the cleanup loop.)
+
+Before invoking build, snapshot untracked files so the post-build delta can be computed:
+```bash
+FEATURE_FILE_LIST="/tmp/autodev_feature_files_${K}.txt"
+git ls-files --others --exclude-standard | sort > /tmp/autodev_untracked_before_${K}.txt
+```
+
+Invoke `Skill("mastermind:build")` with a detailed brief:
+
+```
+Brief must include:
+  - Feature title and user_value
+  - Concrete spec: inputs, outputs, API surface, CLI flags, UI, etc.
+  - Entry point file(s) to create or modify — list every path explicitly
+  - Acceptance criteria (3–5 testable outcomes)
+  - What NOT to touch (blast radius guard)
+  - Test requirements: at minimum one unit test and one integration test
+```
+
+Wait for build to complete before Phase B.
+
+After build returns, determine which files this feature owns for staging in Phase D. The **build brief's declared file list** is the primary source of truth — it names every path the build was allowed to touch. Supplement it with only the files the build created that were not untracked before:
+
+```bash
+# 1. Write declared paths from the build brief (one per line)
+printf '%s\n' \
+  "path/to/file1.ts" \
+  "path/to/file2.ts" \
+  "tests/path/to/test1.ts" \
+  > "$FEATURE_FILE_LIST"
+
+# 2. Append any newly untracked files the build produced beyond the brief
+#    (comm -13: lines only in file2 — i.e., new since the before-snapshot)
+comm -13 \
+  /tmp/autodev_untracked_before_${K}.txt \
+  <(git ls-files --others --exclude-standard | sort) \
+  >> "$FEATURE_FILE_LIST"
+```
+
+Store build start in memory:
+```
+mcp__monomind__memory_store(
+  content: "autodev newfeature building: <title>",
+  namespace: "mastermind:autodev",
+  tags: ["autodev", "newfeature", "building"]
+)
+```
+
+#### Phase B — Review Until Clean
+
+Before starting the review loop, snapshot current working tree state (both unstaged and untracked) so the post-review delta is scoped only to this phase:
+```bash
+git diff --name-only | sort > /tmp/autodev_unstaged_before_B_${K}.txt
+git ls-files --others --exclude-standard | sort > /tmp/autodev_untracked_before_B_${K}.txt
+```
+
+Inline review loop (same cap as improvement mode: max 5 iterations):
+
+```
+review_iter = 0
+LOOP:
+  review_iter += 1
+  If review_iter > 5: EXIT LOOP (cap reached — log warning, continue)
+
+  Invoke Skill("mastermind:review") with:
+    prompt: "Review the new feature just built: <title>. Verify: correctness, edge cases,
+             tests present and passing, no regressions, security, API consistency."
+    brain_context: brain_context
+    mode: auto
+    (do NOT pass --tillend)
+
+  If review returns zero findings: EXIT LOOP (clean)
+  Else: auto-fix findings; continue loop
+```
+
+After the review loop exits, append only files this phase introduced (delta from before-snapshot), covering both modified tracked files and newly created untracked files:
+```bash
+# Tracked files modified by review (delta only — excludes files modified by prior features)
+comm -13 \
+  /tmp/autodev_unstaged_before_B_${K}.txt \
+  <(git diff --name-only | sort) \
+  >> "$FEATURE_FILE_LIST"
+# Untracked files created by review (e.g. new test files)
+comm -13 \
+  /tmp/autodev_untracked_before_B_${K}.txt \
+  <(git ls-files --others --exclude-standard | sort) \
+  >> "$FEATURE_FILE_LIST"
+sort -u "$FEATURE_FILE_LIST" -o "$FEATURE_FILE_LIST"
+```
+
+#### Phase C — Documentation
+
+Before generating documentation, snapshot current working tree state:
+```bash
+git diff --name-only | sort > /tmp/autodev_unstaged_before_C_${K}.txt
+git ls-files --others --exclude-standard | sort > /tmp/autodev_untracked_before_C_${K}.txt
+```
+
+After review is clean (or capped), generate documentation for the feature.
+
+Invoke `Skill("mastermind:content")` with:
+```
+prompt: "Document the new feature '<title>' that was just built.
+         Generate:
+         1. A concise user-facing description (2-3 sentences) for README or CHANGELOG
+         2. API/CLI/function-level docstrings for every new public symbol
+         3. A usage example (runnable snippet or command)
+         4. Any caveats or known limitations
+         Write documentation inline into the relevant files — do not create standalone doc files
+         unless the project already has a /docs directory pattern."
+brain_context: brain_context
+mode: auto
+```
+
+If `Skill("mastermind:content")` raises a "skill not found" or "skill unavailable" error, write the documentation directly:
+- Add docstrings to all new public functions/classes/commands
+- Update README.md (Features section or equivalent) with one bullet describing the feature
+- Add a CHANGELOG entry under `[Unreleased]` only if `CHANGELOG.md` already exists in the project (check with `git ls-files CHANGELOG.md`); do NOT create it from scratch
+
+After Phase C completes, append only files this phase introduced (delta from before-snapshot):
+```bash
+comm -13 \
+  /tmp/autodev_unstaged_before_C_${K}.txt \
+  <(git diff --name-only | sort) \
+  >> "$FEATURE_FILE_LIST"
+comm -13 \
+  /tmp/autodev_untracked_before_C_${K}.txt \
+  <(git ls-files --others --exclude-standard | sort) \
+  >> "$FEATURE_FILE_LIST"
+sort -u "$FEATURE_FILE_LIST" -o "$FEATURE_FILE_LIST"
+```
+
+#### Phase D — Delivery
+
+(`$FEATURE_FILE_LIST` is the variable set at the top of Phase A: `/tmp/autodev_feature_files_${K}.txt`.)
+
+Stage only the files declared for this feature. Do NOT commit — leave that to the user.
+
+```bash
+# Load file list, filtering blank lines, into an array (bash 4+)
+mapfile -t _feat_files < <(grep -v '^[[:space:]]*$' "$FEATURE_FILE_LIST" 2>/dev/null)
+
+# Guard: skip if file list is missing or contains only blank lines
+if [ ${#_feat_files[@]} -eq 0 ]; then
+  echo "[autodev:newfeature] Warning: Feature ${K} file list is empty — build may be a no-op. Skipping stage."
+else
+  # Stage only this feature's declared files. Exclude secrets/env.
+  for f in "${_feat_files[@]}"; do
+    case "$f" in
+      .env|*.env|secrets*|credentials*) echo "Skipping sensitive file: $f" ;;
+      *) git add -- "$f" 2>/dev/null || echo "Warning: could not stage $f" ;;
+    esac
+  done
+
+  # Guard: detect no-op build (files listed but nothing actually changed)
+  # Quoted array expansion handles filenames with spaces correctly
+  if git diff --cached --quiet -- "${_feat_files[@]}"; then
+    echo "[autodev:newfeature] Warning: Feature ${K} staged no changes — build produced no diff."
+  else
+    # Show what this feature contributed to the staged set (scoped to feature files only)
+    git diff --cached --stat -- "${_feat_files[@]}"
+  fi
+fi
+```
+
+Log delivery:
+```
+[autodev:newfeature] Feature <K>/<newfeature_count> staged: <title>
+Status: <staged|no-op|skipped>   # staged=changes in index, no-op=built but no diff, skipped=build failed/empty list
+Review: <clean|capped>
+Files changed: <git diff --cached --stat output scoped to this feature's files>
+```
+
+Track the outcome for FP-End: `feature_status[K] = staged|no-op|skipped`, `feature_review[K] = clean|capped|n/a`.
+
+Suggest a commit message (print it, do not run it):
+```
+feat(<entry_point_module>): <title>
+
+<user_value>
+
+Delivered by mastermind:autodev --newfeature
+```
+
+Store completion in memory:
+```
+mcp__monomind__memory_store(
+  content: "autodev newfeature built: <title> — <user_value>. Status: <status>.",
+  namespace: "mastermind:autodev",
+  tags: ["autodev", "newfeature", "built", status]
+)
+```
+
+---
+
+### FP-End — Summary
+
+After all `newfeature_count` features complete, clean up tmp files and print a summary table:
+
+```bash
+# Re-assign K to the final feature count (each shell block is independent per Phase A rule)
+K=<final_newfeature_count>   # substitute the actual integer, e.g. K=3
+# Clean up per-feature tmp files
+for k in $(seq 1 $K); do
+  rm -f /tmp/autodev_feature_files_${k}.txt \
+        /tmp/autodev_untracked_before_${k}.txt \
+        /tmp/autodev_unstaged_before_B_${k}.txt \
+        /tmp/autodev_untracked_before_B_${k}.txt \
+        /tmp/autodev_unstaged_before_C_${k}.txt \
+        /tmp/autodev_untracked_before_C_${k}.txt
+done
+```
+
+Count staged features: `staged_count = count of features where feature_status[K] == "staged"`.
+
+Print summary table (Status: `staged|no-op|skipped`; Review: `clean|capped|n/a`):
+
+```
+╔══════════════════════════════════════════════════════════════╗
+║  autodev --newfeature  |  Session Summary                    ║
+╠══════════════════════════════════════════════════════════════╣
+║  Feature  │  Title                  │  Status   │  Review    ║
+╠═══════════╪═════════════════════════╪═══════════╪════════════╣
+║  1/<newfeature_count>  │  <title>   │  staged   │  clean     ║
+║  2/<newfeature_count>  │  <title>   │  skipped  │  n/a       ║
+║  ...                                                         ║
+╚══════════════════════════════════════════════════════════════╝
+Staged: <staged_count>/<newfeature_count> features. Run `git commit` to finalize.
+```
+
+---
+
+## Loop: For Each Improvement (1 through `count`)
+
+Run this sequence `count` times. After each completed improvement, increment the counter and continue — do not stop between improvements unless `mode = confirm` and the user declines.
+
+### Phase 1 — Research
+
+**Goal:** Understand the project deeply enough to pick the single best improvement to build right now.
+
+Gather context from at minimum these sources — run all in parallel:
+
+```bash
+# Git log: what has been worked on recently
+git log --oneline -20
+
+# Project structure
+find . -maxdepth 3 -type f \( -name "*.ts" -o -name "*.js" -o -name "*.py" -o -name "*.go" -o -name "*.md" \) \
+  ! -path "*/node_modules/*" ! -path "*/.git/*" ! -path "*/dist/*" | head -60
+
+# Package/project metadata
+cat package.json 2>/dev/null || cat pyproject.toml 2>/dev/null || cat go.mod 2>/dev/null || true
+
+# README
+cat README.md 2>/dev/null | head -80 || true
+
+# Existing issues/todos (common patterns)
+grep -rn "TODO\|FIXME\|HACK\|XXX\|BUG\|PERF\|DEBT" --include="*.ts" --include="*.js" --include="*.py" \
+  --include="*.go" --include="*.md" . 2>/dev/null | grep -v node_modules | grep -v dist | head -30
+```
+
+Also call (parallel):
+- `mcp__monomind__monograph_god_nodes` — find high-centrality files (architectural hotspots)
+- `mcp__monomind__monograph_suggest` with query `"most impactful improvement"` — graph-guided suggestions
+- `mcp__monomind__memory_search` with query `"autodev built"`, namespace `"mastermind:autodev"`, limit 20 — to exclude already-built improvements from candidates
+
+If a `focus` hint was provided, bias all research toward that topic.
+
+**Output of research phase:**
+Produce a ranked list of 3–5 candidate improvements. For each candidate:
+- `title`: one-line name
+- `type`: feature | bugfix | refactor | performance | dx | security | test
+- `why`: one sentence — why this helps the project most
+- `feasibility`: high | medium (skip low — don't pick things that need days)
+- `blast_radius`: files/modules affected (estimate)
+
+### Phase 2 — Selection
+
+Pick the **single best** candidate using this priority order:
+1. Highest feasibility
+2. Lowest blast radius for the value delivered
+3. Aligns with `focus` hint (if given)
+4. Not something already in git log from the last 20 commits
+
+If `mode = confirm`: show the ranked list and selection, wait for user approval before proceeding.
+
+If `mode = auto`: proceed immediately with the top pick.
+
+Log the selection:
+```
+[autodev] Improvement <N>/<count>: <title> (<type>)
+Rationale: <why>
+Files affected: <blast_radius>
+```
+
+Store selection in memory so it's not re-picked next iteration:
+```
+mcp__monomind__memory_store(
+  content: "autodev built: <title> — <why>",
+  namespace: "mastermind:autodev",
+  tags: ["autodev", "built", type]
+)
+```
+
+### Phase 3 — Build
+
+Invoke `Skill("mastermind:build")` with:
+- `prompt`: detailed implementation brief for the selected improvement
+- `brain_context`: the loaded brain context
+- `project_name`: `$(basename "$PWD")`
+- `mode`: `auto`
+- `board_id`: the autodev board (create once at startup, reuse)
+
+The brief passed to build MUST include:
+- What to build (concrete spec, not vague)
+- Which files to touch
+- Acceptance criteria (2–4 testable outcomes)
+- What NOT to change (blast radius guard)
+
+Wait for `mastermind:build` to complete before proceeding to Phase 4.
+
+### Phase 4 — Review Until Clean
+
+After build completes, enter an **inline** review loop (no ScheduleWakeup — must stay within this session):
+
+```
+review_iter = 0
+LOOP:
+  review_iter += 1
+  If review_iter > 5: EXIT LOOP (cap reached)
+  
+  Invoke Skill("mastermind:review") with:
+    prompt: "Review the changes just made for: <title>. Check correctness, tests, edge cases, security."
+    brain_context: brain_context
+    mode: auto
+    (do NOT pass --tillend — that would schedule a ScheduleWakeup and break out of autodev)
+  
+  If review returns zero findings: EXIT LOOP (clean)
+  Else: findings are auto-fixed by the reviewer; continue loop
+```
+
+**Critical:** Do not invoke the review command via `/mastermind:review --tillend ...`. That syntax schedules ScheduleWakeup continuations which would terminate the current autodev session. Always invoke inline with `Skill("mastermind:review")` only.
+
+If the cap (5 iterations) is reached without a clean pass, log a warning and continue to Phase 5 — don't block the outer loop.
+
+### Phase 5 — Log Completion
+
+```
+[autodev] Improvement <N>/<count> complete: <title>
+Status: clean | capped (review issues remain)
+```
+
+Update brain with what was built:
+```
+mcp__monomind__memory_store(
+  content: "autodev completed improvement <N>: <title>. Type: <type>. Status: <status>.",
+  namespace: "mastermind:autodev",
+  tags: ["autodev", "completed", type, status]
+)
+```
+
+If `N < count`: log `[autodev] Moving to improvement <N+1>/<count>...` and repeat from Phase 1.
+
+---
+
+## Standalone Execution
+
+1. Extract flags (leading integer for count, --newfeature N, --focus, --auto/--confirm)
+2. Load brain context via _protocol.md Brain Load Procedure (namespace: `autodev`)
+3. Create monotask board:
+   ```bash
+   project_name="${project_name:-$(basename "$PWD")}"
+   board_id=$(monotask board create "autodev" --json 2>/dev/null | jq -r '.id // empty')
+   ```
+4. **If `--newfeature` was parsed:** run the Feature Pipeline (FP-0 through FP-End) and skip the improvement loop entirely.
+   **Otherwise:** run the Loop section above for each improvement.
+5. At end: follow _protocol.md Brain Write Procedure (namespace: `autodev`)
+
+---
+
+## Tillend Integration
+
+**`--newfeature` is incompatible with `--tillend`.** Both the command and skill strip `--tillend` and emit a warning when both are present. This is because the Feature Pipeline has a fundamentally different termination model (a fixed shortlist of N features, not an open-ended "until exhausted" loop), and the FP-0 discovery + FP-End staging structure does not map onto the per-wakeup session model that `--tillend` assumes.
+
+For the standard improvement loop: when `--tillend` wraps autodev, each ScheduleWakeup wakeup runs a **full autodev session** (all `count` improvements). Memory deduplication (Phase 1 searches `"autodev built"`) prevents already-built improvements from being re-picked in subsequent wakeups.
+
+- `count` controls improvements **per session** (default: 1)
+- `--tillend` keeps scheduling new sessions until an empty round
+- An empty round occurs when Phase 1 research finds no viable new candidates (all feasible work is already in memory)
+
+The loop terminates naturally when the project has no more improvements worth building — not based on hitting a fixed count across wakeups.
+
+---
+
+## Safety Guards
+
+- Never delete files unless the improvement or feature explicitly requires it
+- Never modify `.env`, secrets, credentials
+- Never commit (only stage) — leave commit to the user
+- If Phase A (build) fails or produces no output: skip this feature, log the skip, continue to next feature
+- If Phase B (review) skill is unavailable or throws: log a warning, treat as capped (review issues may remain), continue to Phase C
+- If Phase C (documentation) skill is unavailable: fall back to inline documentation; if that also fails, log and continue to Phase D
+- If Phase D (delivery) `git add` fails on multiple files: log each failure, stage what succeeded, continue
+- If all research candidates are infeasible: ask the user for a focus direction instead of guessing
+- **`--newfeature` specific:**
+  - Features must be genuinely new — reject any candidate that is just a refactor or bugfix of existing code
+  - Do not create standalone `.md` doc files unless the project already has a `/docs` directory
+  - Stage only the files declared in the build brief for each feature (written to `$FEATURE_FILE_LIST` in Phase A), supplemented only by newly untracked files the build produced (detected via before/after `git ls-files` delta) — never use `git add --all` in feature mode
+  - Each feature's staged set is driven by its build brief's declared file list, making it deterministic and cross-feature safe
+  - `newfeature_count > 10` is caught at parse time and capped to 10 before any pipeline runs
+  - `--tillend` combined with `--newfeature` is rejected at parse time with a clear warning

package/.claude/skills/mastermind/createorg.md

@@ -128,6 +128,7 @@ Ask the user (or infer from prompt) for the optional Paperclip-style fields:
       "agent_type": "<subagent_type slug from mapping table>",
       "responsibilities": ["<1-3 bullet responsibilities>"],
       "reports_to": "<role id or null>",
+      "skills": [],
       "adapter_config": {
         "model": "<claude model id>",
         "max_tokens": 8192
@@ -226,18 +227,32 @@ mkdir -p .monomind/orgs
 Write the confirmed org config as JSON using `jq` to guarantee valid encoding:
 
 ```bash
-# Build the config JSON from the confirmed org plan and write atomically
-# (Claude constructs the full jq expression from the confirmed roles/edges in Step 5)
+# Build the config JSON from the confirmed org plan and write atomically.
+# Set shell variables from the confirmed plan before running this block:
+#   governance_policy  — "auto" | "board" | "strict"  (from Step 4)
+#   budget_tokens_val  — integer or 0 for unlimited     (from Step 4)
+#   ceo_adapter        — model id string                (from Step 4)
 jq -n \
   --arg name "$org_name" \
   --arg goal "$goal" \
   --arg topology "$topology" \
   --argjson roles "$roles_json" \
   --argjson communication "$communication_json" \
+  --arg gov_policy "${governance_policy:-auto}" \
+  --argjson budget_tokens "${budget_tokens_val:-0}" \
+  --arg ceo_adapter "${ceo_adapter:-claude-sonnet-4-6}" \
   '{name:$name,goal:$goal,mode:"daemon",topology:$topology,
     created_at:(now|todate),roles:$roles,communication:$communication,
+    governance:{policy:$gov_policy,approvals_file:(".monomind/orgs/"+$name+"-approvals.json")},
     board_space:$name,board_name:"org-tasks",
-    run_config:{checkpoint_interval_min:30,max_concurrent_agents:6,memory_namespace:("org:"+$name)}}' \
+    run_config:{
+      checkpoint_interval_min:30,
+      max_concurrent_agents:6,
+      memory_namespace:("org:"+$name),
+      budget_tokens:$budget_tokens,
+      alert_threshold:0.8,
+      ceo_adapter:$ceo_adapter
+    }}' \
   > "${orgJson}.tmp" && mv "${orgJson}.tmp" "$orgJson"
 ```
 
@@ -296,14 +311,15 @@ curl -s -X POST "${CTRL_URL}/api/mastermind/event" \
     '{type:"domain:complete",session:$session,domain:"ops",status:"complete",
       org:$orgName,goal:$goal,roles_count:$rolesCount,ts:(now*1000|floor)}')" || true
 
-# org:create — so handleOrgEvent routes it to the Orgs panel event log
+# org:create — so handleOrgEvent routes it to the Orgs panel event log and SSE triggers list refresh
 curl -s -X POST "${CTRL_URL}/api/mastermind/event" \
   -H "Content-Type: application/json" \
   -d "$(jq -cn \
     --arg session "$session_id" \
     --arg org "$orgName" \
     --arg goal "$goal_val" \
-    '{type:"org:create",session:$session,org:$org,goal:$goal,ts:(now*1000|floor)}')" || true
+    --arg proj "$(pwd)" \
+    '{type:"org:create",session:$session,org:$org,goal:$goal,project:$proj,ts:(now*1000|floor)}')" || true
 ```
 
 ---