Agent Skills: Secrets Management

Secrets Management

API keys and credentials require careful handling throughout their lifecycle.

ikigai Application

API keys (OpenAI, Anthropic, etc.):

• Store in config file with 0600 permissions
• Load once at startup, hold in memory
• Never log, never include in error messages
• Never embed in source code or commits

Memory handling:

• Scrub secrets from memory when done: explicit_bzero(key, len)
• Avoid strdup() for secrets (can't track copies)
• Keep secret lifetime short and scoped

Config file security:

// Check permissions before reading
struct stat st;
if (stat(path, &st) == 0 && (st.st_mode & 077) != 0) {
    return ERR(ctx, SECURITY, "Config file permissions too open");
}

Never expose:

• In logs or debug output
• In error messages shown to user
• In core dumps (prctl(PR_SET_DUMPABLE, 0))
• Via environment to child processes

Review red flags: Secrets in printf/logging, strdup on credentials, missing permission checks.