johba/disinto
1
0
Fork 0
Code Issues 4 Pull requests Projects Releases Packages Wiki Activity Actions

refactor: planner maintains AGENTS.md instead of STATE.md

Browse source 
STATE.md was a machine-generated system description that was always
worse than the human-curated AGENTS.md. Killed STATE.md entirely.

Phase 1: Reviews recent git history against AGENTS.md, suggests
updates via PR to keep the file tree, conventions, and architecture
descriptions current.

Phase 2: Gap analysis — compares AGENTS.md + VISION.md + open issues,
creates backlog issues for missing capabilities.
This commit is contained in:
openhands 2026-03-15 15:13:34 +00:00
parent e1101894d6
commit 679c62e7cb
1 changed files with 95 additions and 118 deletions
Download patch file Download diff file Expand all files Collapse all files

213
planner/planner-agent.sh
View file

@ -1,10 +1,10 @@
#!/usr/bin/env bash #!/usr/bin/env bash
# ============================================================================= # =============================================================================
# planner-agent.sh — Rebuild STATE.md from git history, then gap-analyse # planner-agent.sh — Keep AGENTS.md current, then gap-analyse against VISION.md
# #
# Two-phase planner run: # Two-phase planner run:
# Phase 1: Rebuild STATE.md from git log + closed issues (compact snapshot) # Phase 1: Review recent git history, suggest AGENTS.md updates via PR
# Phase 2: Compare STATE.md vs VISION.md, create backlog issues for gaps # Phase 2: Compare AGENTS.md vs VISION.md, create backlog issues for gaps
# #
# Usage: planner-agent.sh (no args — uses env vars from .env / env.sh) # Usage: planner-agent.sh (no args — uses env vars from .env / env.sh)
# ============================================================================= # =============================================================================
@ -19,7 +19,7 @@ source "$FACTORY_ROOT/lib/env.sh"
LOG_FILE="$SCRIPT_DIR/planner.log" LOG_FILE="$SCRIPT_DIR/planner.log"
CLAUDE_TIMEOUT="${CLAUDE_TIMEOUT:-3600}" CLAUDE_TIMEOUT="${CLAUDE_TIMEOUT:-3600}"
MARKER_FILE="${PROJECT_REPO_ROOT}/.last-planner-sha" MARKER_FILE="${PROJECT_REPO_ROOT}/.last-planner-sha"
STATE_FILE="${PROJECT_REPO_ROOT}/STATE.md" AGENTS_FILE="${PROJECT_REPO_ROOT}/AGENTS.md"
VISION_FILE="${PROJECT_REPO_ROOT}/VISION.md" VISION_FILE="${PROJECT_REPO_ROOT}/VISION.md"
log() { echo "[$(date -u +%Y-%m-%dT%H:%M:%S)Z] $*" >> "$LOG_FILE"; } log() { echo "[$(date -u +%Y-%m-%dT%H:%M:%S)Z] $*" >> "$LOG_FILE"; }
@ -39,139 +39,127 @@ if [ -f "$MARKER_FILE" ]; then
if git cat-file -e "$LAST_SHA" 2>/dev/null; then if git cat-file -e "$LAST_SHA" 2>/dev/null; then
GIT_RANGE="${LAST_SHA}..HEAD" GIT_RANGE="${LAST_SHA}..HEAD"
else else
log "WARNING: marker SHA ${LAST_SHA:0:7} not found, using 30-day window" log "WARNING: marker SHA ${LAST_SHA:0:7} not found, using 7-day window"
first_sha="" first_sha=$(git log --format=%H --after='7 days ago' --reverse 2>/dev/null | head -1) || true
first_sha=$(git log --format=%H --after='30 days ago' --reverse 2>/dev/null | head -1) || true GIT_RANGE="${first_sha:-HEAD~20}..HEAD"
GIT_RANGE="${first_sha:-HEAD~30}..HEAD"
fi fi
else else
log "No marker file, using 30-day window" log "No marker file, using 7-day window"
first_sha=$(git log --format=%H --after='30 days ago' --reverse 2>/dev/null | head -1) || true first_sha=$(git log --format=%H --after='7 days ago' --reverse 2>/dev/null | head -1) || true
GIT_RANGE="${first_sha:-HEAD~30}..HEAD" GIT_RANGE="${first_sha:-HEAD~20}..HEAD"
fi fi
GIT_LOG=$(git log "$GIT_RANGE" --oneline --no-merges 2>/dev/null || true) GIT_LOG=$(git log "$GIT_RANGE" --oneline --no-merges 2>/dev/null || true)
MERGE_LOG=$(git log "$GIT_RANGE" --oneline --merges 2>/dev/null || true)
COMMIT_COUNT=$(echo "$GIT_LOG" | grep -c '.' || true) COMMIT_COUNT=$(echo "$GIT_LOG" | grep -c '.' || true)
log "Range: $GIT_RANGE ($COMMIT_COUNT commits)" log "Range: $GIT_RANGE ($COMMIT_COUNT commits)"
if [ "$COMMIT_COUNT" -eq 0 ] && [ -f "$STATE_FILE" ]; then # ── File tree (for context on what exists) ───────────────────────────────
log "No new commits since last run — skipping STATE.md rebuild" FILE_TREE=$(find . -maxdepth 3 -type f \( -name '*.sol' -o -name '*.ts' -o -name '*.sh' -o -name '*.md' -o -name '*.json' \) \
# Still run gap analysis (vision or issues may have changed) ! -path '*/node_modules/*' ! -path '*/.git/*' ! -path '*/out/*' ! -path '*/cache/*' ! -path '*/dist/*' \
2>/dev/null | sort | head -200)
# ── Phase 1: AGENTS.md maintenance ──────────────────────────────────────
if [ "$COMMIT_COUNT" -eq 0 ]; then
log "No new commits since last run — skipping AGENTS.md review"
else else
# ── Phase 1: Rebuild STATE.md ────────────────────────────────────────── log "Phase 1: reviewing AGENTS.md against recent changes"
log "Phase 1: rebuilding STATE.md"
CURRENT_STATE="" CURRENT_AGENTS=""
[ -f "$STATE_FILE" ] && CURRENT_STATE=$(cat "$STATE_FILE") [ -f "$AGENTS_FILE" ] && CURRENT_AGENTS=$(cat "$AGENTS_FILE")
# Read project docs for high-level understanding if [ -z "$CURRENT_AGENTS" ]; then
PROJECT_DOCS="" log "No AGENTS.md found — skipping phase 1"
for doc in AGENTS.md docs/PRODUCT-TRUTH.md docs/ARCHITECTURE.md docs/UX-DECISIONS.md; do else
[ -f "${PROJECT_REPO_ROOT}/${doc}" ] && \ # Files changed in this range
PROJECT_DOCS="${PROJECT_DOCS} FILES_CHANGED=$(git diff --name-only "$GIT_RANGE" 2>/dev/null | sort -u || true)
### ${doc}
$(cat "${PROJECT_REPO_ROOT}/${doc}")
"
done
# Fetch recently closed issues for context PHASE1_PROMPT="You maintain AGENTS.md for the ${PROJECT_NAME} repository. AGENTS.md is the primary onboarding document — it describes the project's architecture, file layout, conventions, and how to work in the codebase.
CLOSED_ISSUES=$(codeberg_api GET "/issues?state=closed&type=issues&limit=30&sort=updated&direction=desc" 2>/dev/null | \
jq -r '.[] | "#\(.number) \(.title)"' 2>/dev/null || true)
PHASE1_PROMPT="You are maintaining STATE.md — a compact factual snapshot of what ${PROJECT_NAME} currently is and does. ## Current AGENTS.md
${CURRENT_AGENTS}
## Project Documentation (read for context — understand the system before writing) ## Recent commits (since last review)
${PROJECT_DOCS:-"(no docs found)"} ${GIT_LOG}
## Current STATE.md ## Files changed
${CURRENT_STATE:-"(empty — create from scratch)"} ${FILES_CHANGED}
## New commits since last snapshot ## Repository file tree (top 3 levels)
${GIT_LOG:-"(none)"} ${FILE_TREE}
## Merge commits
${MERGE_LOG:-"(none)"}
## Recently closed issues
${CLOSED_ISSUES:-"(none)"}
## Task ## Task
Update STATE.md to describe what the project IS right now — its architecture, capabilities, and current state. Review AGENTS.md against the recent changes. Output an UPDATED version of AGENTS.md that:
- Reflects any new directories, scripts, tools, or conventions introduced by the recent commits
- Removes or updates references to things that were deleted or renamed
- Keeps the existing structure, voice, and level of detail — this is a human-curated document, preserve its character
- Does NOT add issue/PR references — AGENTS.md is timeless documentation, not a changelog
- Does NOT rewrite sections that haven't changed — preserve the original text where possible
### Writing style If AGENTS.md is already fully up to date and no changes are needed, output exactly: NO_CHANGES
- Describe the SYSTEM, not the changes. Think: 'What would a new developer need to know?'
- Lead with what the project does at a high level, then drill into subsystems
- Group related capabilities together (e.g. all evolution stuff in one bullet)
- Issue/PR references (#42) are good for traceability but should SUPPORT descriptions, not replace them
- Bad: 'evolve.sh: auto-incrementing results directory (#752), cleans stale tmpdirs (#750)'
- Good: 'Evolution pipeline runs perpetually on a dedicated VPS, breeding Push3 optimizer bytecode through mutation, crossover, and elitism against a revm fitness evaluator'
- No dates, no changelog framing, no 'fixed X' or 'added Y' language
- No more than 25 bullet points — be concise and architectural
- First bullet should be a one-line project description
Your response must start with '- ' on the very first character. No summary, no meta-commentary, no preamble, no markdown fences, no 'Here is...' or 'I have...'. ONLY the bullet list. Any response not starting with '- ' will be rejected." Otherwise, output the complete updated AGENTS.md content (not a diff, the full file).
Do NOT wrap the output in markdown fences. Start directly with the file content."
PHASE1_OUTPUT=$(timeout "$CLAUDE_TIMEOUT" claude -p "$PHASE1_PROMPT" \ PHASE1_OUTPUT=$(timeout "$CLAUDE_TIMEOUT" claude -p "$PHASE1_PROMPT" \
--model sonnet \ --model sonnet \
2>/dev/null) || { --dangerously-skip-permissions \
log "ERROR: claude exited with code $? during phase 1" 2>/dev/null) || {
exit 1 log "ERROR: claude exited with code $? during phase 1"
} # Update marker even on failure to avoid re-processing same range
echo "$HEAD_SHA" > "$MARKER_FILE"
exit 1
}
if [ -z "$PHASE1_OUTPUT" ]; then if echo "$PHASE1_OUTPUT" | grep -q "^NO_CHANGES$"; then
log "ERROR: empty output from phase 1" log "AGENTS.md is up to date — no changes needed"
exit 1 elif [ -n "$PHASE1_OUTPUT" ]; then
fi # Write updated AGENTS.md and create PR
TEMP_FILE=$(mktemp "${AGENTS_FILE}.XXXXXX")
printf '%s\n' "$PHASE1_OUTPUT" > "$TEMP_FILE"
mv "$TEMP_FILE" "$AGENTS_FILE"
# Strip any non-bullet preamble (Sonnet sometimes narrates before the bullets) if ! git diff --quiet "$AGENTS_FILE" 2>/dev/null; then
PHASE1_OUTPUT=$(echo "$PHASE1_OUTPUT" | sed -n '/^- /,$p') branch_name="chore/planner-agents-$(date -u +%Y%m%d)"
git checkout -B "$branch_name" 2>/dev/null
git add "$AGENTS_FILE"
git commit -m "chore: planner update AGENTS.md" --quiet 2>/dev/null
git push -f origin "$branch_name" --quiet 2>/dev/null || { log "ERROR: failed to push $branch_name"; git checkout "${PRIMARY_BRANCH}" 2>/dev/null; }
git checkout "${PRIMARY_BRANCH}" 2>/dev/null
# Restore AGENTS.md to master version after branch push
git checkout "$AGENTS_FILE" 2>/dev/null || true
# Validate output has bullet points # Create or update PR
if [ -z "$PHASE1_OUTPUT" ] || ! echo "$PHASE1_OUTPUT" | head -1 | grep -q '^- '; then EXISTING_PR=$(codeberg_api GET "/pulls?state=open&limit=50" 2>/dev/null | \
log "ERROR: phase 1 output contains no bullet list" jq -r --arg branch "$branch_name" '.[] | select(.head.ref == $branch) | .number' | head -1)
exit 1 if [ -z "$EXISTING_PR" ]; then
fi PR_RESPONSE=$(codeberg_api POST "/pulls" \
"{\"title\":\"chore: planner update AGENTS.md\",\"head\":\"${branch_name}\",\"base\":\"${PRIMARY_BRANCH}\",\"body\":\"Automated AGENTS.md update based on recent commits (${COMMIT_COUNT} changes since last review).\"}" \
# Atomic write 2>/dev/null)
TEMP_STATE=$(mktemp "${STATE_FILE}.XXXXXX") PR_NUM=$(echo "$PR_RESPONSE" | jq -r '.number // empty')
printf '%s\n' "$PHASE1_OUTPUT" > "$TEMP_STATE" if [ -n "$PR_NUM" ]; then
mv "$TEMP_STATE" "$STATE_FILE" log "Created PR #${PR_NUM} for AGENTS.md update"
matrix_send "planner" "📋 PR #${PR_NUM}: planner update AGENTS.md (${COMMIT_COUNT} commits reviewed)" 2>/dev/null || true
# Commit STATE.md via PR (master is protected) else
if ! git diff --quiet "$STATE_FILE" 2>/dev/null; then log "ERROR: failed to create PR"
branch_name="chore/planner-state-$(date -u +%Y%m%d)" fi
git checkout -B "$branch_name" 2>/dev/null else
git add "$STATE_FILE" log "Updated existing PR #${EXISTING_PR}"
git commit -m "chore: planner rebuild STATE.md" --quiet 2>/dev/null fi
git push -f origin "$branch_name" --quiet 2>/dev/null || { log "ERROR: failed to push $branch_name"; git checkout "${PRIMARY_BRANCH}" 2>/dev/null; return 1; }
git checkout "${PRIMARY_BRANCH}" 2>/dev/null
# Create or update PR (filter by head ref — Codeberg's head param is unreliable)
EXISTING_PR=$(codeberg_api GET "/pulls?state=open&limit=50" 2>/dev/null | jq -r --arg branch "$branch_name" '.[] | select(.head.ref == $branch) | .number' | head -1)
if [ -z "$EXISTING_PR" ]; then
PR_RESPONSE=$(codeberg_api POST "/pulls" "{\"title\":\"chore: planner rebuild STATE.md\",\"head\":\"${branch_name}\",\"base\":\"${PRIMARY_BRANCH}\",\"body\":\"Automated STATE.md rebuild from git history + closed issues.\"}" 2>/dev/null)
PR_NUM=$(echo "$PR_RESPONSE" | jq -r '.number // empty')
if [ -n "$PR_NUM" ]; then
log "Created PR #${PR_NUM} for STATE.md update"
matrix_send "planner" "📋 PR #${PR_NUM}: planner rebuild STATE.md" 2>/dev/null || true
else else
log "ERROR: failed to create PR" log "AGENTS.md diff was empty after write — no PR needed"
fi fi
else
log "Updated existing PR #${EXISTING_PR}"
fi fi
fi fi
# Update marker # Update marker
echo "$HEAD_SHA" > "$MARKER_FILE" echo "$HEAD_SHA" > "$MARKER_FILE"
log "Phase 1 done — STATE.md rebuilt ($(wc -l < "$STATE_FILE") lines)" log "Phase 1 done"
fi fi
# ── Phase 2: Gap analysis ─────────────────────────────────────────────── # ── Phase 2: Gap analysis ───────────────────────────────────────────────
log "Phase 2: gap analysis" log "Phase 2: gap analysis"
CURRENT_STATE=$(cat "$STATE_FILE" 2>/dev/null || true) AGENTS_CONTENT=$(cat "$AGENTS_FILE" 2>/dev/null || true)
VISION="" VISION=""
[ -f "$VISION_FILE" ] && VISION=$(cat "$VISION_FILE") [ -f "$VISION_FILE" ] && VISION=$(cat "$VISION_FILE")
@ -190,34 +178,27 @@ fi
OPEN_SUMMARY=$(echo "$OPEN_ISSUES" | jq -r '.[] | "#\(.number) [\(.labels | map(.name) | join(","))] \(.title)"' 2>/dev/null || true) OPEN_SUMMARY=$(echo "$OPEN_ISSUES" | jq -r '.[] | "#\(.number) [\(.labels | map(.name) | join(","))] \(.title)"' 2>/dev/null || true)
# Fetch vision-labeled issues specifically
VISION_ISSUES=$(echo "$OPEN_ISSUES" | jq -r '.[] | select(.labels | map(.name) | index("vision")) | "#\(.number) \(.title)\n\(.body)"' 2>/dev/null || true)
PHASE2_PROMPT="You are the planner for ${CODEBERG_REPO}. Your job: find gaps between the project vision and current reality. PHASE2_PROMPT="You are the planner for ${CODEBERG_REPO}. Your job: find gaps between the project vision and current reality.
## VISION.md (human-maintained goals) ## VISION.md (human-maintained goals)
${VISION} ${VISION}
## STATE.md (current project snapshot) ## AGENTS.md (current project state — the ground truth)
${CURRENT_STATE} ${AGENTS_CONTENT}
## Vision-labeled issues (goal anchors)
${VISION_ISSUES:-"(none)"}
## All open issues ## All open issues
${OPEN_SUMMARY} ${OPEN_SUMMARY}
## Task ## Task
Identify gaps — things implied by VISION.md that are neither reflected in STATE.md nor covered by an existing open issue. Identify gaps — things implied by VISION.md that are neither reflected in AGENTS.md nor covered by an existing open issue.
For each gap, output a JSON object (one per line, no array wrapper): For each gap, output a JSON object (one per line, no array wrapper):
{\"title\": \"action-oriented title\", \"body\": \"problem statement + why it matters + rough approach\", \"depends\": [list of blocking issue numbers or empty]} {\"title\": \"action-oriented title\", \"body\": \"problem statement + why it matters + rough approach\", \"depends\": [list of blocking issue numbers or empty]}
## Rules ## Rules
- Max 5 new issues — focus on highest-leverage gaps only - Max 5 new issues — focus on highest-leverage gaps only
- Do NOT create issues for things already in STATE.md (already done) - Do NOT create issues for things already described in AGENTS.md (already done)
- Do NOT create issues that overlap with ANY existing open issue, even partially - Do NOT create issues that overlap with ANY existing open issue, even partially
- Do NOT create issues about vision items, tech-debt, or in-progress work
- Each title should be a plain, action-oriented sentence - Each title should be a plain, action-oriented sentence
- Each body should explain: what's missing, why it matters for the vision, rough approach - Each body should explain: what's missing, why it matters for the vision, rough approach
- Reference blocking issues by number in depends array - Reference blocking issues by number in depends array
@ -228,6 +209,7 @@ Output ONLY the JSON lines (or NO_GAPS) — no preamble, no markdown fences."
PHASE2_OUTPUT=$(timeout "$CLAUDE_TIMEOUT" claude -p "$PHASE2_PROMPT" \ PHASE2_OUTPUT=$(timeout "$CLAUDE_TIMEOUT" claude -p "$PHASE2_PROMPT" \
--model sonnet \ --model sonnet \
--dangerously-skip-permissions \
2>/dev/null) || { 2>/dev/null) || {
log "ERROR: claude exited with code $? during phase 2" log "ERROR: claude exited with code $? during phase 2"
exit 1 exit 1
@ -240,21 +222,18 @@ if echo "$PHASE2_OUTPUT" | grep -q "NO_GAPS"; then
fi fi
# ── Create issues from gap analysis ────────────────────────────────────── # ── Create issues from gap analysis ──────────────────────────────────────
# Find backlog label ID
BACKLOG_LABEL_ID=$(codeberg_api GET "/labels" 2>/dev/null | \ BACKLOG_LABEL_ID=$(codeberg_api GET "/labels" 2>/dev/null | \
jq -r '.[] | select(.name == "backlog") | .id' 2>/dev/null || true) jq -r '.[] | select(.name == "backlog") | .id' 2>/dev/null || true)
CREATED=0 CREATED=0
while IFS= read -r line; do while IFS= read -r line; do
[ -z "$line" ] && continue [ -z "$line" ] && continue
# Skip non-JSON lines
echo "$line" | jq -e . >/dev/null 2>&1 || continue echo "$line" | jq -e . >/dev/null 2>&1 || continue
TITLE=$(echo "$line" | jq -r '.title') TITLE=$(echo "$line" | jq -r '.title')
BODY=$(echo "$line" | jq -r '.body') BODY=$(echo "$line" | jq -r '.body')
DEPS=$(echo "$line" | jq -r '.depends // [] | map("#\(.)") | join(", ")') DEPS=$(echo "$line" | jq -r '.depends // [] | map("#\(.)") | join(", ")')
# Add dependency section if present
if [ -n "$DEPS" ] && [ "$DEPS" != "" ]; then if [ -n "$DEPS" ] && [ "$DEPS" != "" ]; then
BODY="${BODY} BODY="${BODY}
@ -262,10 +241,7 @@ while IFS= read -r line; do
${DEPS}" ${DEPS}"
fi fi
# Create issue
CREATE_PAYLOAD=$(jq -nc --arg t "$TITLE" --arg b "$BODY" '{title:$t, body:$b}') CREATE_PAYLOAD=$(jq -nc --arg t "$TITLE" --arg b "$BODY" '{title:$t, body:$b}')
# Add label if we found the backlog label ID
if [ -n "$BACKLOG_LABEL_ID" ]; then if [ -n "$BACKLOG_LABEL_ID" ]; then
CREATE_PAYLOAD=$(echo "$CREATE_PAYLOAD" | jq --argjson lid "$BACKLOG_LABEL_ID" '.labels = [$lid]') CREATE_PAYLOAD=$(echo "$CREATE_PAYLOAD" | jq --argjson lid "$BACKLOG_LABEL_ID" '.labels = [$lid]')
fi fi
@ -273,6 +249,7 @@ ${DEPS}"
RESULT=$(codeberg_api POST "/issues" -d "$CREATE_PAYLOAD" 2>/dev/null || true) RESULT=$(codeberg_api POST "/issues" -d "$CREATE_PAYLOAD" 2>/dev/null || true)
ISSUE_NUM=$(echo "$RESULT" | jq -r '.number // "?"' 2>/dev/null || echo "?") ISSUE_NUM=$(echo "$RESULT" | jq -r '.number // "?"' 2>/dev/null || echo "?")
log "Created #${ISSUE_NUM}: ${TITLE}" log "Created #${ISSUE_NUM}: ${TITLE}"
matrix_send "planner" "📋 Gap issue #${ISSUE_NUM}: ${TITLE}" 2>/dev/null || true
CREATED=$((CREATED + 1)) CREATED=$((CREATED + 1))
[ "$CREATED" -ge 5 ] && break [ "$CREATED" -ge 5 ] && break