Cognitive Spawn
Avvia i watcher cognitivi in background (inhibit, error, cleanup). Sono daemon bash che fanno tail del transcript della sessione corrente + check periodici sul filesystem, scrivono alert in ~/.claude-cognitive-alerts/, e il PostToolUse hook li consegna al modello prima del prossimo tool call. Invoca ad inizio sessione di lavoro non triviale, dopo /clear, o quando vuoi monitoring real-time attivo.…
Install
$ agentstack add skill-tommasobbianchi-claude-cognitive-skills-cognitive-spawn Open-source listing — not yet scanned by AgentStack. Follow the source repository for install instructions.
Security review
⚠ Flagged1 finding(s); flagged for manual review. · v0.1.0 How review works →
- • Prompt-injection patterns
- • Secret / credential exfiltration
- • Dangerous shell & filesystem operations
- • Untrusted network calls
- • Known-malicious package signatures
- high Destructive filesystem operation.
What it can access
- ✓ Network access No
- ✓ Filesystem access No
- ✓ Shell / process execution No
- ✓ Environment & secrets No
- ✓ Dynamic code execution No
From automated source analysis of v0.1.0. “Used” means the capability is present in the source — more access means more to trust, not that it’s unsafe.
About
Cognitive-Spawn — sentinelle in background
Principio: i sottosistemi di controllo girano in processi separati dal main, osservano transcript+filesystem, e injettano alert via hook PostToolUse prima del prossimo tool call mutativo. Vedi /cognitive-flow per il modello completo.
Come avviarli
Identifica il transcript della sessione corrente e lancia:
TRANSCRIPT=$(ls -t ~/.claude/projects/$(pwd | sed 's|/|-|g')/*.jsonl | head -1)
~/.claude/cognitive-watchers/spawn.sh start "$TRANSCRIPT" "$PWD"Lo script avvia 3 daemon (PID files in ~/.claude-cognitive-alerts/.pids/):
| Watcher | Cosa fa | Sleep | |---|---|---| | inhibit | tail del transcript JSONL; detect comandi Bash ad alto blast radius (rm -rf, git push --force, deploy, scp, drop table) | 2 s | | error | tail del transcript; detect tool_result con is_error=true | 2 s | | cleanup | check periodico del working tree del cwd + /tmp residui | 60 s |
Come usarli
Niente — sono passivi. Quando producono un alert, il file appare in ~/.claude-cognitive-alerts/.alert. Al prossimo tool call mutativo (Bash, Edit, Write, Task), il hook PostToolUse registrato in ~/.claude/settings.json li legge, formatta su stderr, exit 2 → il modello riceve l'alert come system-reminder e DEVE addressarlo prima di proseguire.
Verifica / stop
~/.claude/cognitive-watchers/spawn.sh status # quali girano + alert pendenti
~/.claude/cognitive-watchers/spawn.sh stop # termina tuttiAuto-start a session start (opzionale)
Per averli sempre attivi senza invocazione manuale, aggiungi un hook SessionStart in ~/.claude/settings.json che lancia spawn.sh start con il transcript della sessione (variabile passata dal JSON stdin del hook).
File correlati
- watcher scripts:
~/.claude/cognitive-watchers/{inhibit,error,cleanup}-watcher.sh - orchestratore:
~/.claude/cognitive-watchers/spawn.sh - hook delivery:
~/.claude/cognitive-watchers/post-tool-hook.sh - dir alert:
~/.claude-cognitive-alerts/ - log daemon:
~/.claude-cognitive-alerts/.logs/ - skill compagne:
/cognitive-stop,/cognitive-flow,/inhibit,/cleanup-check,/error-check
Source & license
This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.
- Author: tommasobbianchi
- Source: tommasobbianchi/claude-cognitive-skills
- License: MIT
Install and usage instructions live in the source repository linked above.
Reviews
No reviews yet — be the first.
Write a review
Versions
- v0.1.0 Imported from the upstream source.