Drops the additive-union semantics that resurrected deliberate deletions
across the fleet (see feedback_memory_sync_destructive_ok.md and the
2026-06-01 consolidation that came back the next morning).
New behavior:
* file in REPO, not in PROFILE -> copy REPO -> PROFILE (unchanged)
* file in PROFILE, not in REPO -> DELETE from PROFILE (was: copy back)
* file in BOTH, identical -> no-op
* file in BOTH, differ -> overwrite PROFILE (was: log conflict)
Safety: aborts if the repo has <5 .md files (guards against a broken
repo wiping the profile store).
Test plan verified on GURU-BEAST-ROG:
* dry-run + apply matched (2 copies + 10 overwrites + 0 deletes)
* idempotent re-run = 79 identical, 0 ops
* self-check memory category PASS
* git status .claude/memory/ clean (script touched profile only)
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
jq -r '.stdout' returns the literal string "null" when the API field is JSON
null, causing the RESULT: grep to fail and fire a false drift alert. Fixes:
- Use `.stdout // empty` so null becomes empty string
- Add FINAL_ST tracking; treat non-terminal status as INFRA-ERROR, not drift
- Increase poll window from 20x4s=80s to 30x4s=120s for slow commands
- Read .stderr and .exit_code; include them in the no-RESULT diagnostic
Live check 2026-06-02: KSTEENBB2025 is PASS (today's alert was a false positive).
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
compute_output_path now parses .gitmodules and, for a project scope whose
dir is a submodule (guru-rmm, guru-connect, youtube-sync-docker), falls
back to the MAIN repo root session-logs/ per convention. Non-submodule
projects (gururmm-agent, dataforth-dos) unchanged.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Reconstructs session logs from Claude Code transcripts when a session
crashes or is closed before /save. Two entry points:
- /recover <uuid|latest> : manual, Claude-reviewed reconstruction
- detect_orphaned_sessions.py : scheduled scan that auto-builds logs for
substantive, unsaved, not-yet-recovered transcripts (banner-marked
RECOVERED-UNVERIFIED), commits them, and posts a #bot-alerts FYI.
recover_session.py is the shared engine: Python extracts the verbatim
command/config/reference timeline; Ollama drafts prose-only narrative.
Machine-local ledger (.claude/state/) prevents reprocessing. Reviewed:
git add scoped to own files, ledger written only after successful push,
per-uuid idempotency, --max cap for unattended runs.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
memory-dream: read-only memory lint/consolidation analyzer (index, backlinks,
stale refs, dup clusters, profile drift); additive-only --apply-safe, all
merges/deletes are proposals. sync-memory.sh: additive repo<->harness-profile
union (no delete/overwrite, conflicts surfaced), wired to a SessionStart hook.
Migrates the useful profile-only memories into the synced repo store.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Require SecurityCenter2 productState RTP-enabled bit before treating a
registered AV as active (lapsed/disabled AV no longer suppresses the
critical Defender finding), and tighten the Datto fallback to AV/EDR
services only — excluding Datto RMM/Backup/Workplace/Continuity/File so
non-AV Datto products can't masquerade as antivirus. Fix misleading comment.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The probe flagged ACG's own MSP tooling (ScreenConnect/ConnectWise Control,
Splashtop, Syncro, Datto RMM, Datto EDR/AV) as CRITICAL "foreign agent" and
flagged Defender-off as CRITICAL even when a 3rd-party AV had legitimately
disabled it. Now: allowlisted tools emit an INFO "expected ACG tooling"
finding (genuinely-foreign tools still CRITICAL); Defender-off is downgraded
to INFO only when a 3rd-party AV is active. JSON contract + grading unchanged.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
PowerShell ConvertTo-Json collapses a single-element array into a bare
object (or, for string arrays, a bare string). The runner iterated/joined
several facts.* fields, so single-volume / single-NIC / single-admin
machines silently dropped the Fixed Volumes table and errored the network
adapter, local-administrator, and installed-software-diff lines.
Fix jq-side in the runner (backward-compatible with already-written
immutable baselines; PS1 untouched per the todo decision) using
`if type=="array" then . elif .==null then [] else [.] end` at:
volumes, network_adapters (+ inner ip/dns), local_administrators, and
installed_software (both sides of the diff). Verified with synthetic
single-element JSON and a multi-element no-regression check.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
/rmm diagnose: dispatches a Windows security/health probe to a newly onboarded
agent, grades RED/AMBER/GREEN, writes an immutable per-client baseline
(clients/<slug>/onboarding-baselines/), diffs vs prior, and alerts CRITICALs to
#dev-alerts. Probe is PS5.1/ASCII/SYSTEM-safe, never-abort, base64 chunked upload
around the agent command-size cap. Code-reviewed (no blockers); folded in
immutability guard, severity-independent finding ids, Defender-unknown sentinel,
expanded competitor/backup detection.
First baselines captured: Rednour FRONTDESKRECEPT + LEGALASST (both RED - prior
MSP ScreenConnect/Splashtop/Syncro still live; LEGALASST OS EOL).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Corrected the 2026-05-28 SmartBadge fix on KSTEENBB2025: the older Datto
Workplace Desktop v8 had been left in place (diverged from the fleet, which
runs Datto Workplace v10.53.4 / Workplace2). Removed v8, installed v10,
aligned the SmartBadge _CC add-in + CLSID to the EVO-X1 reference, and cleared
Kristin's stuck per-user LoadBehavior=2.
- ksteen-smartbadge-verify.ps1: PASS/FAIL verdict vs fleet reference
- ksteen-smartbadge-fix.ps1: machine + per-user remediation
- check-ksteen-smartbadge.sh: daily runner (RMM -> verdict -> #bot-alerts,
coord message to Mike on drift); driven by a 7-day scheduled task on GURU-5070
- wiki: agents table, dual-Workplace SmartBadge known issue + fleet standard,
2026-05-28/29 history
Syncro #32339. Coord todo 4a5b09b3 (watch expires 2026-06-05).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
sanitize_json() called `python3` unconditionally, but on ACG Windows boxes
the Microsoft Store python3 alias is disabled and `py` is the launcher
(feedback_python_windows). When `python3` was missing the function silently
returned empty, and the surrounding `result_safe='{"messages":[]}'` default
dropped every unread coord message — no error, no warning, no toast.
Now prefers identity.json's `.python.command` (set during machine onboarding,
matching the pattern other scripts already use), falls back to
`command -v python3 || command -v py || command -v python`, and if no Python
is available falls back to `tr -d '\000-\037'` so jq can still parse — lossy
on real \n/\t in string fields but keeps messages visible instead of dropping
them.
Phase 2 migration complete:
sync.sh:
- Read Python command from identity.json first (.python.command)
- Fall back to auto-detection for legacy machines
- Eliminates per-session detection overhead
syncro.md:
- Read Ollama endpoint from identity.json (.ollama.endpoint // .ollama.fallback)
- Read Python command from identity.json (.python.command)
- Both sections have legacy fallbacks with detection
- Eliminates 2-second curl probe on every write operation
- Updated day-of-week verification code example
- Updated Ollama draft call section
Impact: All scripts now read machine-specific config from identity.json
(populated by migrate-identity.sh). Faster, explicit, offline-safe.
The script auto-detected PYTHON_CMD but then hardcoded `python3` for the
JSON write (exit 127 on Windows where only `py` exists), and passed a Git
Bash POSIX path (/d/...) to native Python (FileNotFoundError). Fixes:
- use "$PYTHON_CMD" instead of hardcoded python3
- convert IDENTITY_PATH via `cygpath -m` for the interpreter (no-op elsewhere)
Verified on GURU-5070: identity.json migrated correctly (py, windows/amd64,
localhost Ollama, qwen3:8b).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
- Auto-detects Python command, platform, architecture
- Probes Ollama (local vs remote)
- Sets prose_model based on machine (qwen3:8b for GURU-5070, else qwen3:14b)
- Tested on Mikes-MacBook-Air: all fields populated correctly
Ready for coord rollout to all machines.
- Updated sync.sh to read claudetools_root from identity.json
- Updated syncro.md skill to use identity.json for repo path
- Updated CLAUDE.md onboarding to include claudetools_root field
- Eliminates cross-architecture path detection issues
- Fallback to git rev-parse for legacy machines
Each machine sets claudetools_root during onboarding, just like vault_path.
Migration 20260526_150000 adds nullable due_at datetime column. Model, schemas
(create/update/response), and sync.sh display updated. Sync output now shows
due:YYYY-MM-DDTHH:MM alongside any todo with a due date.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
New coord_todos table and API endpoints (GET/POST/PUT/DELETE /api/coord/todos)
supporting manual and auto-created items, sub-tasks via parent_id, and inclusive
for_user/for_machine filters (OR-null) for sync/save display. sync.sh Phase 7
now shows pending todos grouped by project after every sync. CLAUDE.md documents
auto-creation behavior for unresolved follow-up. Web/email pricing doc updated:
block time rate clarified, INKY reference removed, dates updated.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Problem: macOS hostname command returns 'Mikes-MacBook-Air.local' but
coord messages are addressed to 'Mikes-MacBook-Air/claude-main'. Hook
script was querying for wrong session ID, so messages never displayed.
Fix: Strip .local suffix using bash parameter expansion before building
session ID.
Result: Coord messages now display correctly on macOS machines.
Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
Phase 1a used `git submodule foreach`, which only visits initialized
submodules — a fresh machine silently skipped population while the script
still reported success. It now explicitly inits + populates each submodule
declared in .gitmodules (inheriting credentials from the parent origin URL
so non-interactive init authenticates; .gitmodules stays credential-free),
then advances each to its remote branch tip. Reporting reflects the real
configured count.
Also removes the orphaned `projects/solverbot` gitlink (mode 160000),
committed without a .gitmodules entry, which made `git submodule status`
throw `fatal: no submodule mapping`. solverbot's own Gitea repo is untouched.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
- Added wiki change detection and categorization in sync.sh
- Shows articles by type (clients/projects/systems/patterns/meta)
- Displays status (added/modified/deleted) and counts
- Updated sync.md and save.md documentation
Removed 5 tracked files whose names were mangled Windows paths (C:\ProgramData\...,
D:\claudetools\.claude\current-mode, and version-bump script fragments) — the
illegal :/\ chars were stored as Unicode Private-Use-Area substitutes (U+F03A, U+F00A).
Added purge_garbled_paths() to sync.sh: before each 'git add -A', any UNTRACKED path
containing control chars, ':', '\', or PUA U+E000-U+F8FF bytes is removed and not
staged, so this cruft never reaches Gitea again. Tracked content is never auto-deleted.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
sync.sh: replace `git diff-index --quiet HEAD --` with
`[ -n "$(git status --porcelain)" ]` in both the main-repo (Phase 1) and
vault change-detection, so brand-new untracked files are no longer silently
skipped (the bug Howard hit 2026-04-17). Mark project_sync_script_bug.md
RESOLVED.
.gitignore: exclude the datto BSOD dumps (6 MB zip + 48 MB extracted) so the
detection fix doesn't sweep 54 MB of binaries into the repo.
memory: finish the add_line_item reconciliation — drop legacy "time entry" /
timer-billable framing from feedback_syncro_labor_type and
feedback_syncro_warranty_product (and their index lines); the product-selection
rules themselves are unchanged.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Add Phase 5b — one-way repo -> global copy of .claude/commands/*.md after
the pull, so the CLI always loads the latest skills. Idempotent (only
new/changed files) and soft-failing (never aborts a sync). Fixes the drift
where the global command set lagged the repo (e.g. /syncro was ~3 weeks
stale, and feature-request/forum-post/inject-standards/shape-spec were
missing globally) because the automated script skipped the doc's Phase 3.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Add .claude/scripts/post-bot-alert.sh — reusable, soft-failing Discord
poster that reads the bot token from the SOPS vault (bot-token.sops.yaml,
credentials.bot_token) with a .env fallback, so it works from any machine.
Wire it into the /syncro skill: a Hard Rules pointer, a billing-workflow
step (17), and a "Post to #bot-alerts" reference section with the message
format and ticket/invoice/customer link mapping (computerguru.syncromsp.com).
Scoped to write ops (create/update/close/comment/bill/customer); reads post
nothing. Best-effort — never fails the Syncro write it follows.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
HOWARD-HOME installs to C: not D:, causing sync.sh to fail with
"Not in a git working tree" on that machine.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
The UserPromptSubmit hook requires .claude/current-mode to determine work mode
and gate coordination lock checks. This file is machine-local (gitignored) but
had no initialization logic for fresh clones, causing hooks to fail.
Changes:
- check-messages.sh: Added auto-creation logic with "general" as default
- CLAUDE.md: Documented auto-initialization behavior
- ONBOARDING.md: Added machine-local configuration section
- session-logs/2026-05-19-session.md: Documented investigation and fix
Impact:
- Fixes coordination hooks on all machines
- Prevents first-clone hook failures
- No manual setup required
- Backwards compatible
Resolves: "cood hook seems to be broken on all my machines"
Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
Hook was querying only to_session=HOSTNAME/claude-main, missing messages
addressed to the short alias (e.g. "howard"). Now reads identity.json for
the alias and queries both, merging results before display.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Python open() can't read MSYS-style paths (/c/claudetools/...).
Fix: try jq first (handles Unix paths cleanly on all platforms),
fall back to Python with cygpath -m conversion to mixed Windows paths.
Matches the same fix already applied to get-token.sh.
Bug reported by Howard (HOWARD-HOME, 2026-04-21).
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
sync.sh: after pull, scan changed session logs for "## Note for" /
"## Message for" sections and print them in a highlighted block
before the sync summary. Forces attention on inter-team messages.
CLAUDE.md: document mandatory behavior — cross-user notes displayed
at top of response with full content, action items addressed before
continuing.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
