Talon ClawTalon Studio v1.0.2-beta
Skills
The playbooks agents can search and pin: a curated cyber-security library plus your own house skills, attached to each agent by choice.
The skills library
Talon Claw → Skills (/claw/skills) browses the playbooks your agents can pin into a run. It holds two kinds:
The vendored cyber-skills library. A broad blue-team catalogue that Claw filters to exactly 10 in-scope subdomains: threat-detection, threat-hunting, threat-intelligence, soc-operations, security-operations, endpoint-security, cloud-security, network-security, identity-security and identity-access-management. Off-environment playbooks (honeypots, packet crafting, disk and memory forensics) never surface. Library content is third-party and treated as untrusted data when shown to a model.
House skills. Your own playbooks, listed first. An admin clicks New house skill and scopes it to the Organisation or the current Workspace; a workspace skill is also written to the workspace repository as Talon/skills/<name>.md.
How agents use skills
Every built-in agent carries a first-party house procedure, its always-on method, and ships with no vendored skills selected. An admin chooses the playbooks each agent may use in its Knowledge tab (Talon Claw → Agents → Configure); organisation house skills are opt-in the same way, while a workspace's house skills are always eligible for runs in that workspace. During a run an agent finds relevant playbooks with search_skills and reads one with load_skill.
Automatic pinning. At dispatch, up to 2 of the agent's eligible skills scoring at least 6 for relevance are pinned into its context, each capped at 6,000 characters. Only skills in the "now" and "soon" tiers are pinned.
Manual load_skill. A skill the agent loads mid-run is capped at 16,000 characters: the full playbook, not just the pinned excerpt.