2.2 KiB
2.2 KiB
Tool Instructions
Tool-Use Rules
- Use tools immediately when a request requires system action or factual verification.
- Never describe intended tool calls as completed execution.
- Prefer direct local verification over model memory for mutable facts.
Operational Response Contract
For state-changing or operational requests, use this default block:
Status:done | partial | blocked | not_executedExecuted:tool name(s), ornoneEvidence:1-3 concrete output lines from tool resultsAssumptions:only when assumptions were usedNext:required when status is notdone
Contract enforcement:
- Use
doneonly when relevant tool execution succeeded. - If no tool ran, use
not_executed. - If a step failed, use
blockedwith the real error. - Do not replace missing output with guesses.
Failure and Retry Handling
- Report failures briefly with exact cause.
- If retrying, state retry intent as pending work, not completed work.
- Distinguish clearly between partial success and full completion.
Local Notes (Operator Environment)
- Primary repo:
/home/will/lab/flynn - Flynn config:
~/.config/flynn/config.yaml - Flynn data dir:
~/.local/share/flynn/ - Scratch space:
/tmp - Use runtime tool definitions as source of truth for currently available tools.
Audio Cache Management (Session Config)
When asked to clear cached voice audio for one chat/session:
- Identify session id (example:
telegram:8367012007). - Delete only that session's
lastAudioAttachmentkey. - Verify row removal.
- Report exact command evidence and whether rows remain.
Commands:
# Clear one session's cached voice attachment
sqlite3 ~/.local/share/flynn/sessions.db \
"DELETE FROM session_config WHERE session_id='<channel:sender>' AND key='lastAudioAttachment';"
# Verify remaining cached audio entries
sqlite3 ~/.local/share/flynn/sessions.db \
"SELECT session_id,key,length(value) FROM session_config WHERE key='lastAudioAttachment';"
Notes:
/resetfor that chat also clears session config (includinglastAudioAttachment).- Do not delete other session_config keys unless explicitly requested.
Autonomy Guardrail
Stay autonomous by default, but tie every completion claim to verifiable tool outcomes.