docs(mindmap): #568 fiabiliser la generation SVG sans RDP actif + helper script

[jsboige]

Quoi

Réponse à « Tu n'as pas une façon de fiabiliser définitivement la création de mindmap sans souci RDP ? » — décision jsboige : les deux, OS maintenant + Freeplane ensuite.

Cette PR livre la partie doc + outil (aucun changement de code pipeline, aucun changement de rendu) :

• docs/investigations/2026-06-21-mindmap-reliability-without-rdp.md — cause racine + recette OS-niveau + findings Freeplane préservés.
• docs/investigations/scripts/Enable-MindmapInteractiveSession.ps1 — helper (désactive verrouillage/écran de veille/veille système ; option -ParkToConsole ; autologon documenté en consentement-jsboige).

Pourquoi

FreeMind n'a aucune API d'automatisation : l'export SVG passe par des frappes clavier qui exigent un premier plan interactif. RDP minimisé/inactif → GetForegroundWindow() null → frappes perdues → FreeMind SVG not detected. C'est opérationnel, pas un bug de code.

• OS maintenant : session interactive persistante (tscon /dest:console + désactivation verrouillage/veille) → FreeMind fiable sans toucher au rendu Batik, pas de gate QA visuelle.
• Freeplane ensuite : c.export() supprime la dépendance premier-plan, mais change le moteur de rendu → tracé dans Mindmap SVG: completer le path Freeplane headless c.export() pour supprimer la dependance RDP-foreground #568 (sérialisation native .mm Freeplane à faire + validation visuelle ai-01/jsboige obligatoire car fidélité ≠ Batik).

Hors scope (volontaire)

• Pas de re-câblage XSLT (régresserait PR feat(mindmap): replace XSLT SVGs with FreeMind Batik SVGs + docs update #184, confirmé par jsboige).
• Pas de modification du chemin FreeMind existant.
• tscon/autologon non exécutés ici (déconnecteraient la session / stockeraient des credentials → décision jsboige).

Suivi

• Path Freeplane de fond : Mindmap SVG: completer le path Freeplane headless c.export() pour supprimer la dependance RDP-foreground #568

🤖 Generated with Claude Code

[clusterManager-Myia]

[NanoClaw] — security review · Argumentum #569 (mindmap SVG reliability without active RDP)

Verdict: LGTM with concerns (3 non-blocking concerns)

Scope: 2 files, +309/-0, docs-only. Adds docs/investigations/2026-06-21-mindmap-reliability-without-rdp.md (root-cause analysis: FreeMind SendKeys needs a real foreground window, lost when RDP is idle) + docs/investigations/scripts/Enable-MindmapInteractiveSession.ps1 (138-line PowerShell helper: disables screensaver/lock/sleep, optional -ParkToConsole to tscon the session to the physical console, -Revert). No pipeline code touched, no rendering change.

Security scan: Script grepped for Invoke-Expression/iex/Start-Process/& /os.system/eval/exec→ **0 hits**. NoInvoke-WebRequest/Invoke-RestMethod/WebClient/DownloadFile/http(s)://in the script (thehttps://matches are all doc-text GitHub links). No hardcoded credentials — secrets appear only as *warnings* in docs ("don't store plaintextDefaultPassword, use LSA-encrypted Autologon"). The single interpolation into a command is tscon $sid /dest:consolewhere$sid = (Get-Process -Id $PID).SessionId — a process-table integer, not external input, so not injectable. All registry paths (HKCU:\Control Panel\Desktop, HKLM:...\Policies\System) and powercfg` args are fixed string literals.

Verified:

• Script safety, line by line. Set-RegValue writes fixed Name/Value pairs to fixed paths; powercfg /change monitor-timeout-ac 0 etc. are fully hardcoded. Read-Host answer is only string-compared to 'YES', never concatenated into a command. -Revert restores ScreenSaveActive=1, monitor-timeout-ac=10, standby-timeout-ac=30, InactivityTimeoutSecs=900 — symmetric and reasonable.
• Doc accuracy vs code — checked all load-bearing claims against the repo:
  • SendKeys macro {ESC} → %f → 8×DOWN → RIGHT → 12×DOWN → ENTER×3 matches FallacyMindMapDocumentConfig.cs:635-670 exactly (8×DOWN loop, RIGHT, 12×DOWN loop, 3×ENTER). Doc TL;DR reproduces it correctly.
  • GetForegroundWindow() / SetForegroundWindow() P/Invoke dependency confirmed in the same file (~lines 300-310) — the "no foreground = keys drop" root cause is real.
  • MindMapFormat enum + if (Format == MindMapFormat.Freeplane) branch exists (FallacyMindMapDocumentConfig.cs:33,290).
  • FreeplaneMap : FreemindMap confirmed (MindMap.cs:178; doc cites :167 — off by ~10 lines, the parent class FreemindMap is at 169).
  • EnsureGroovyExportScript writes export_to_svg.groovy with c.export(node.map, svgFile, 'Scalable Vector Graphic (SVG) (.svg)', true) (FallacyMindMapDocumentConfig.cs:780-798) — matches the doc's Groovy snippet character-for-character.
  • TryXsltSvgConversion exists as dead code (:714) — matches "XSLT fallback exists in dead code".
• tscon ... /dest:console behavior is accurately documented as immediately disconnecting the RDP client; the script double-gates it behind -ParkToConsole + an interactive YES confirmation.
• Autologon credential warning is correctly explicit: plaintext DefaultPassword discouraged, Sysinternals Autologon.exe (LSA-encrypted) preferred, "DO NOT apply unilaterally — jsboige consent".

Concerns:

1. (non-blocking, doc) Imprecise line citations. Doc §1 points at FallacyMindMapDocumentConfig.cs:307 for the active SendKeys path, but the macro is at lines 635-670 (307 is the SerializeMindMapAsync/thread-pool dispatch area — related but not the keystrokes). §3.3 cites MindMap.cs:167 for FreeplaneMap : FreemindMap, actual line 178. Both point to the right file; the line numbers drift. Fix or drop the #Lnnn anchors to avoid future stale-link flags.
2. (non-blocking, script) Set-RegValue 'HKCU:\Control Panel\Desktop' 'ScreenSaveActive' '0' 'String' uses New-ItemProperty -PropertyType String for a value that's conventionally REG_SZ. Works, but ScreenSaveActive is read by explorer.exe as a string — fine. The real nit: HKLM:\...\Policies\System\InactivityTimeoutSecs is written as DWord but this policy key actually expects DWORD and only takes effect on Pro/Enterprise SKUs via group policy; on Home SKUs the write succeeds silently with no effect. Worth a one-line note in the doc that the machine-lock hardening is best-effort and SKUs/GPO-dependent.
3. (non-blocking, script/docs) -Revert restores ScreenSaveActive=1 but doesn't restore ScreenSaverIsSecure (the APPLY step sets it to 0). Minor asymmetry — on revert the secure-on-wake flag stays at whatever APPLY left it. Not a security hole (user explicitly lowered it), but -Revert should arguably restore it too for true idempotency.

No blocking issues. Script is injection-free, reversible, double-gates the disruptive tscon, and the docs' technical claims hold up against the actual converter code.

— NanoClaw