Summary
inference-gateway/cli#761 adds a new user-facing config key, agent.skills.max_chars, and changes two system-prompt behaviors. The CLI configuration reference needs to cover them.
What changed:
- New key
agent.skills.max_chars (default 4000, 0 disables, env INFER_AGENT_SKILLS_MAX_CHARS): caps the rendered AVAILABLE SKILLS block in the agent system prompt. Skills past the budget are listed by name only (never a truncated description) and remain invocable via /<name>.
- The injected memory index (
memory.max_chars) now truncates at a line boundary instead of mid-entry.
- The system prompt now carries a date-only
Current date: line instead of a minute-resolution timestamp, so provider prompt-prefix caching survives across turns.
Example config:
agent:
skills:
enabled: true
max_chars: 4000 # 0 disables the cap
Source PR: inference-gateway/cli#761
Acceptance Criteria
Summary
inference-gateway/cli#761 adds a new user-facing config key,
agent.skills.max_chars, and changes two system-prompt behaviors. The CLI configuration reference needs to cover them.What changed:
agent.skills.max_chars(default4000,0disables, envINFER_AGENT_SKILLS_MAX_CHARS): caps the rendered AVAILABLE SKILLS block in the agent system prompt. Skills past the budget are listed by name only (never a truncated description) and remain invocable via/<name>.memory.max_chars) now truncates at a line boundary instead of mid-entry.Current date:line instead of a minute-resolution timestamp, so provider prompt-prefix caching survives across turns.Example config:
Source PR: inference-gateway/cli#761
Acceptance Criteria
agent.skills.max_charsdocumented in the CLI configuration reference (default,0semantics,INFER_AGENT_SKILLS_MAX_CHARSenv override)/<name>