the-agency-ai/the-agency/flag

Flag

Zero-friction capture of observations during work. Flag something now, discuss it later. Flags are stored in the ISCP SQLite DB (outside git) — instant capture from any worktree.

Arguments

• $ARGUMENTS: The message to flag, or a subcommand (list, clear, count, discuss), or --to <agent> <message> to route to a specific agent.

Instructions

Flagging an item (to self)

If $ARGUMENTS is a message (not a subcommand):

bash $CLAUDE_PROJECT_DIR/agency/tools/flag "$ARGUMENTS"

Confirm: "Flagged: [message summary]"

Flagging to a specific agent

If $ARGUMENTS starts with --to:

bash $CLAUDE_PROJECT_DIR/agency/tools/flag --to <agent-address> "message"

The agent address must be fully qualified (repo/principal/agent). The flag appears in the target agent's queue on their next flag list or iscp-check.

Listing the queue

If $ARGUMENTS is list:

bash $CLAUDE_PROJECT_DIR/agency/tools/flag list

Displays all unread flags and marks them as read (Slack-style seen behavior). Three-state lifecycle: unread → read (on list) → processed (on discuss/clear).

Counting items

If $ARGUMENTS is count:

bash $CLAUDE_PROJECT_DIR/agency/tools/flag count

Returns unread count only.

Clearing the queue

If $ARGUMENTS is clear:

bash $CLAUDE_PROJECT_DIR/agency/tools/flag clear

Marks all flags as processed.

Discussing flagged items

If $ARGUMENTS is discuss:

bash $CLAUDE_PROJECT_DIR/agency/tools/flag discuss

Outputs the queue as a numbered 1B1 agenda and marks all as processed. Then invoke /discuss with the agenda items for structured resolution.

Flag Lifecycle

capture → unread → read (on list) → processed (on discuss/clear)

Flags are DB-only — no git payload, instant capture. They're designed for quick observations during work, not formal communication (use dispatches for that).

OFFENDERS WILL BE FED TO THE — CUTE — ATTACK KITTENS!