EXO email-cleanup tasks (Search-UnifiedAuditLog, Get-MessageTrace, inbox rules) kept
401/403-ing per tenant because the Exchange Operator SP was missing the Exchange Admin
directory role — admin consent grants Exchange.ManageAsApp but never the directory role.
onboard-tenant.sh assigns it, but tenants consented before that step / by hand never got
it, and nothing audited for it. Hence the recurring 'next onboarding will fix it' (false
for already-onboarded tenants).
- NEW assign-exchange-role.sh: idempotent role assignment via the authoritative
roleManagement/directory/roleAssignments API (the legacy directoryRoles/members list
reads back unreliably). <domain|--all> + --verify/--dry-run.
- Backfilled the whole fleet (--all): 13 stragglers ASSIGNED, 12 already OK, 20 skipped
(tenant-admin not consented), 0 errors. Safe Site included.
- Standing audit documented (assign-exchange-role.sh --all --verify) + memory so no future
session repeats the empty promise.
- Adds wiki/clients/safesite.md (tenant + 4-source endpoint inventory + investigation).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Live-verified 2026-06-08: Security Investigator + User Manager + Tenant Admin Graph
tiers all consented and reading (subscribedSkus/organization HTTP 200) on
safesitellc.com (71b4e637-...). The reference's 'NO' was stale (last touched 2026-04-20).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Builds the false-positive/true-positive proof the plan requires before the guard can be
promoted to blocking, and fixes the one false-positive it surfaced.
- test-harness-guard.sh: 12-case matrix in a throwaway repo, runs the REAL guard, asserts
WARN/clean for real conflicts/secrets/keys vs legit content (setext underlines, dividers,
docs that mention a marker, encrypted sops, public keys, .example templates).
- harness-guard.sh: conflict rule now requires a real hunk (BOTH ^<<<<<<< AND ^>>>>>>>),
dropping the lone =======$ trigger that false-positived on a 7-char setext underline /
divider. Identical true-positive power (git writes all three markers); FP surface -> 0.
- /self-check: new harness.guard_selftest runs the matrix in an isolated temp repo (read-only
vs the real tree) so guard correctness is continuously proven.
Verified 12/12 pass, true positives intact, real-tree FP surface = 0. FATAL flip (todo
f1c11d0d, on/after 2026-06-22) is now evidence-backed + one-step.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Task 3 leftover. Adds a 'consistency' category to /self-check that catches a standard
drifting back into restating/contradicting the command that owns the rule -- the Syncro
timers failure mode (standard said 'always timer' while /syncro said 'outlier only').
Deterministic half: each manifest.command_standard_links pair's standard must still carry
its defer-to-SSOT pointer (must_reference regex). Lost pointer = WARN. Seeded with
syncro-billing (time-entry-protocol.md -> /syncro). Semantic contradiction pass delegated
to the model in SKILL.md, mirroring check_memory. Verified PASS; negative-tested.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Adds a 'harness' category to /self-check (Task 12, self-check half) so the harness-
optimization gains can't silently regress. All read-only / non-invasive:
- VERSION marker present + not older than manifest.harness.min_version
- skill-registry description budget (sum of all SKILL.md description: fields under
registry_desc_budget_chars) -- the metric that catches Task 5 bloating back
- global deploy targets ~/.claude/skills + ~/.claude/commands populated (Mac-wipe failure)
- harness-guard.sh present + wired into sync.sh
- core scripts parse (bash -n on sync/guard/now-phoenix); now-phoenix.sh emits a valid date
Tunables in baseline/manifest.json 'harness' block. Verified 9/9 PASS; budget WARN
negative-tested at a synthetic over-budget value.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The context-lookup standard + CODING_GUIDELINES still said 'GrepAI First' unconditionally.
Updated both to: wiki first for known-entity facts; GrepAI/Grep-before-read for code+discovery.
Keeps the search-before-read token discipline; removes the wiki overlap. Completes the
positioning fix started in e8a689b0 (all 4 sources now consistent: CORE, EXTENDED, standard,
guidelines).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Resolves the contradiction between CORE (wiki-first) and EXTENDED (which said
'use GrepAI first for any context lookup'). New order: wiki for known entities ->
GrepAI for code call-graphs / discovery / un-compiled detail -> raw reads. Keeps
GrepAI's irreplaceable code-search value; removes the redundant wiki overlap.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Safety prerequisites for the P0 rollout, landed BEFORE any sync.sh change so a bad
harness change cannot strand a node. .claude/harness/VERSION (1.0.0) lets a session
detect partial rollout; .claude/scripts/force-pull-raw.sh is a hook-free git rescue
(dry-run by default; --confirm hard-resets to origin/main, saving prior HEAD to a
recovery branch).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Task 3/3a of the harness-optimization spec. Mike confirmed normal billing uses
add_line_item; timers stay available only for explicit outlier requests, never the
normal loop. Rewrote time-entry-protocol.md to defer to the /syncro command (SSOT for
billing mechanics) and state timers are outlier-only; aligned the command's two
absolute "no timers" lines. Contradiction removed.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Vault sync was failing with "remote: Failed to authenticate user" against
git.azcomputerguru.com. Root cause: Git Credential Manager (first in the
helper chain) shadowed the valid PAT in the store helper with a stale
cached OAUTH_USER JWT.
Fix (machine-local git config, already applied — not in the repo):
- Reset the vault repo credential.helper to store-only (drop inherited GCM).
- Pin azcomputerguru@ in the vault remote URL so store returns the durable
PAT instead of a volatile OAUTH_USER JWT.
Repo change here is documentation only: a feedback memory capturing the
diagnosis + fix, plus an index line in MEMORY.md.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Add .claude/scripts/ff.py, a Firefox browser driver built on Playwright and
the Firefox sibling of the existing cdp.py Chrome driver. It runs a small
background daemon holding one Playwright Firefox page on a persistent profile,
controlled over localhost:9333, with subcommands launch/status/nav/shot/click/
type/eval/console/network/stop. Verified end-to-end (real screenshot, network
and console capture). This is now the preferred browser-automation path because
Mike dislikes Chrome and the claude-in-chrome extension (that connector was
disabled in ~/.claude.json this session - not a repo change).
Add memory reference_ff_firefox_driver.md documenting the driver and an index
line in MEMORY.md. The MEMORY.md change also unavoidably includes a pre-existing
adjacent index line for reference_antigravity_agy_not_headless.md, so that memory
file is bundled in to keep the index consistent.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Amend windows-bootstrap.ps1 with every gap the 2026-06-06 GURU-5070
reinstall exposed, so the next rebuild is clean:
- Phase 7: install python deps into BOTH interpreters (py/3.14 for vault
+ scripts, python/3.12 for the MCP servers). Single-interpreter installs
left ticktick MCP (no httpx/mcp in 3.12) and vault get-field (no PyYAML
in 3.14) dead. Add pyyaml + websocket-client to the baseline libs.
- Phase 3: persist ~\.grok\bin (+ ~\.local\bin, %APPDATA%\npm) to the User
PATH; grok's installer leaves it session-only.
- Phase 6: prime non-interactive git auth (setup-git-auth.sh) so pushes
never hang on a GCM prompt.
- Phase 8: expand to the real 5-model set and add the hydration gotcha so a
populated D:\OllamaModels is never needlessly re-downloaded (~48 GB).
Document all four in machines/guru-5070.md known issues.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Add setup-git-auth.sh: idempotent, fail-silent script that primes the
git credential store from the vault Gitea token, scoped per-repo by the
actual origin host. Only seizes the helper from the prompting GCM
`manager` (leaves Mac osxkeychain alone); fast-path no-op once set.
Wire it into a backgrounded SessionStart hook and set
GIT_TERMINAL_PROMPT=0 / GCM_INTERACTIVE=Never in settings.json env so
no session on any machine can hang on a credential prompt.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Mike's objection to Git for Windows is the constant GCM password
prompts that hang automation/background pushes, not the tool itself.
Document the working fix (repo-local credential.helper=store primed
with the azcomputerguru Gitea API token, GIT_TERMINAL_PROMPT=0) in the
Gitea Agent definition and shared memory.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Rename the machine to the name in the bundle's identity.json (default GURU-5070,
override with -Hostname) when run as admin, with an end-of-run reboot reminder.
Ensures scheduled tasks, coord session IDs, and log attribution line up. RESTORE.md
documents the step.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Add restore-at-risk-work.ps1 and wire it into bootstrap Phase 6. Recreates
local-only WIP rescued to the recovery bundle's at-risk-work/: re-applies the
three guru-rmm stash patches back AS stashes (LIFO order preserved) and drops
the guru-connect tmp-spec018.diff back as its untracked working file. Patches
that won't apply cleanly are reported for manual git apply --3way. Updates
RESTORE.md and the session log with the rescue details.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Add .claude/bootstrap/ (windows-bootstrap.ps1, restore-secrets.ps1,
backup-to-bundle.ps1, RESTORE.md) plus machines/guru-5070.md. Idempotent
11-phase rebuild after a clean Windows reset: winget core tools + .NET/WiX,
protoc, Poppler, Tailscale; restore SOPS age key/SSH/tool-auth/identity from
the E:/F: recovery bundle; clone repos+submodules; set OLLAMA_MODELS/HOST/PROTOC;
detect existing D:\OllamaModels; register scheduled tasks. Includes session log.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
sync.sh now posts a per-machine coord component
(claudetools/git_sync_<MACHINE>) flipped syncing -> idle/degraded around
each run, so the fleet can see who is mid-sync / last sync state. Fully
best-effort: a 3s-capped curl guarded with || true + return 0, emitted
only after the lock is acquired (contention/exit-75 emits nothing), and
finalize captures $? first and returns it so the signal can never change
the sync's real exit code. Reviewed (verified it cannot break sync).
Extract the per-machine concurrency lock from sync.sh into a sourceable
lib (.claude/scripts/sync-lock.sh) plus a `run <cmd>` wrapper that locks
the current repo (same lock-dir basename, so it mutually excludes with
sync.sh in the ClaudeTools repo and self-scopes in any project repo).
sync.sh now sources it (behavior identical — verified by review). /scc
routes its commit+push through the locked, rebase-safe sync.sh (and drops
the bare YYYY-MM-DD-session.md filename for the per-session-unique one).
/checkpoint now stages+commits atomically under the repo lock so a
concurrent session in a shared worktree can't be swept in. Closes the
remaining commit paths that bypassed the lock shipped in 6b0ce9a.
Multiple concurrent Claude sessions (and the scheduled-task sync) were
stepping on each other's git state. sync.sh now takes an atomic mkdir
lock in .git/ around the whole run (stage/commit/fetch/rebase/push +
vault), exits 75 (EX_TEMPFAIL = deferred) on contention instead of
racing, and reclaims stale/dead-owner locks with a re-verify-before-clear
guard (closes two TOCTOU races caught in review). /save now mandates
per-session-unique log filenames (never the bare YYYY-MM-DD-session.md).
Docs updated for the lock + deferred-exit semantics.
Note: git add -A is still the catch-all sweep; full per-session commit
isolation and routing /scc + /checkpoint through the lock are follow-ups.
New `elevate` mode that goes beyond friction to make a UI top-notch and
flags when to redesign rather than patch. references/polish-and-redesign.md
holds 12 heuristics (hierarchy, signature moment, action gravity, narrative,
lonely states, density, rhythm, type, tokens, depth/finish, motion, redesign
triggers) synthesized from three independent model passes (Claude + Gemini +
Grok). Adds an Elevation Index (0-10), a Redesign Urgency score (>=4 leads
with a Structural Audit), and Opportunity-ranked Quick Wins / Elevations /
Redesign Candidates tiers. SKILL.md: command + mode section + extend note.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
- New wiki/systems/ix-server.md: IX web host (172.16.3.10) facts, the
ACG hosted sites table, and a full record of radio.azcomputerguru.com
(Astro static + React 19 islands; source in projects/radio-show/website/;
build npm run build -> dist -> rsync to cPanel doc root).
- index.md: list the new IX systems article.
- radio-show.md: fix the stale "ix-server.md may not exist" backlink.
- memory reference_radio_website.md: add stack detail (React islands,
wavesurfer/fuse, node>=22) + pointer to the new wiki article.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Upgrade the human-flow skill (Gemini-assisted, Claude-reviewed):
- scan.mjs rewritten to AST-based (@babel/parser/traverse) with 4
detectors: unlabeled-icon-button, tiny-target, missing-feedback-props,
click-without-keyboard; regex fallback on parse failure.
- Objective Friction Index (Motor 3.0 / Cognitive 2.5 / Keyboard 2.5 /
Feedback 2.0); 0-10 Human Workflow Score.
- New heuristics: State-Flow Audit, Precision Rail / Fumble Zones,
Restraint-o-Meter (1-5) for the fancy pass.
- `fix` command DISABLED for now (advisory only): the AST generator
reprints whole files and produces noisy diffs; agents apply surgical
fixes from the report. To be revisited with a string-splice editor.
- Add @babel/* deps + package-lock.json.
- Memory: agy review/review-files is NOT actually read-only (wrote files
+ ran npm despite documented plan-mode) — diff after every agy review.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
review/review-files resolve relative paths only against CWD or
$CLAUDETOOLS_ROOT, never a submodule/subdir — so submodule-relative
paths fail with "file not found". Add a [!WARNING] callout to both
SKILL.md files, fix the misleading "absolute or repo-relative" table
wording, and add inline GOTCHA comments at each resolution site in
both scripts. Bitten us repeatedly (latest: GuruConnect review).
Feedback from Mike (Bardach #32387): every Syncro ticket bot-alert needs a
clickable link (https://computerguru.syncromsp.com/tickets/<internal_id>).
post-bot-alert.sh posts raw text, so the URL must be in the message.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Adds explicit Syncro ticket creation section to remediation-tool.md.
Ticket #32387 was created without priority, assignee, or a valid issue type.
Now specifies required fields, valid problem_type values, and an enforcement
checklist to prevent null fields in any POST payload.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
image-analyze: independent second-model vision over OAuth (pins the
gemini-3.1-pro-preview vision model; the default flash-lite router
hallucinates image content) — reads an image via read_file and describes it.
search: Google-grounded live web results with citation URLs (google_web_search).
Both verified working on the keyless Google OAuth. Image GENERATION
(nano-banana) still needs an AI Studio key + extension and stays Grok's lane.
Includes a scoped best-effort output sanitizer for image-analyze (preview
model occasionally leaks reasoning tokens); text/verify/review/search
unchanged. migrate-identity.sh now upgrades the gemini capabilities array.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
If a grok read_file-based review (review/review-files/review-diff) returns
empty (the 0.2.20-style headless tool-gating regression), retry once with
the file(s)/diff embedded inline via the no-tools text path, when content
is under 256KB; otherwise emit a clear skip note. Keeps grok-reads-files as
the default happy path (works on 0.2.22) and degrades gracefully instead of
returning silence. text/verify/raw unchanged; Windows path handling intact.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Live Mailprotector CloudFilter REST client (emailservice.io/api/v1,
Bearer auth via vault msp-tools/mailprotector.sops.yaml). Lists mail-flow
logs and held/quarantined messages across client domains and releases them
(POST messages/{id}/deliver, deliver_many). Read-only by default; every
release/rule-add/config-change gated behind --confirm. Mirrors the
packetdial skill pattern. Built after diagnosing a Dataforth held-outbound
message that never reached ACG.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
role_assigned() sent an unencoded space in the OData $filter
(principalId eq '...'), so the query always failed and the function
always returned false -> onboard-tenant.sh always printed
"MISSING -> ASSIGNING" and relied on the conflict-tolerant POST for
idempotency. Fixed to %20; corrected the stale PIM-misdiagnosis comment.
Verified live against the ACG tenant. Roles still assign correctly;
PRESENT/MISSING reporting is now accurate.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
