AIMFP

Batch-reserve type IDs before writing multiple type definitions. Include definition_json and description for every type.

Restore selected databases from a backup archive, side-by-side rather than in place. Every requested member is extracted to a temporary file FIRST; only once all extractions succeed is anything live touched. The current file is then moved into backups/ as 'replaced-<timestamp>-<filename>' and the restored copy moved into position. Nothing is ever deleted, so the displaced state remains recoverable.

Build the interaction graph for already-cataloged functions: which tracked function calls which. Read-only — returns tuples shaped for add_interactions, which does the writing. Run AFTER catalog_files and catalog_functions. Resolution is conservative and strictest-first (explicit import, then same-file, then module alias, then project-wide unique name), following re-export chains through modules that define no functions of their own. Ambiguous targets are dropped and counted rather than guessed at, because a wrong edge sends future refactors chasing dependencies that do not exist.

Read-only inventory of an existing codebase for adoption into AIMFP tracking. Walks source_directory applying the watchdog's exact exclusion stack (built-in exclusions + watchdog_excluded_* settings + .watchdogignore), extracts signatures, and reports tracked vs untracked. Writes nothing — review the inventory, then register with catalog_files/catalog_functions/catalog_types.

Find directives by type, category, or keyword when you know the domain but not the exact directive name. Multi-word input is split into words; a row matches if ANY word prefix-matches ("disk space free" finds get_free_disk_space). Rows matching more words rank first. Punctuation is ignored, so no FTS syntax is needed or honored. Common stopwords (the, of, and, ...) are ignored unless the query is nothing but stopwords. Keyword results are ordered by name/description relevance.

Search tracked functions by name or purpose with relevance ranking. Multi-word input is split into words; a row matches if ANY word prefix-matches ("disk space free" finds get_free_disk_space). Rows matching more words rank first. Punctuation is ignored, so no FTS syntax is needed or honored. Common stopwords (the, of, and, ...) are ignored unless the query is nothing but stopwords. Results are capped by limit; total_count reports all matches.

Search modules by name, purpose, or description with relevance ranking. Multi-word input is split into words; a row matches if ANY word prefix-matches ("disk space free" finds get_free_disk_space). Rows matching more words rank first. Punctuation is ignored, so no FTS syntax is needed or honored. Common stopwords (the, of, and, ...) are ignored unless the query is nothing but stopwords. Results are capped by limit; total_count reports all matches.

Search past decisions and context by keyword. Multi-word input is split into words; a row matches if ANY word prefix-matches ("disk space free" finds get_free_disk_space). Rows matching more words rank first. Punctuation is ignored, so no FTS syntax is needed or honored. Common stopwords (the, of, and, ...) are ignored unless the query is nothing but stopwords. Returns capped previews by default; read a note in full with get_notes_comprehensive(note_id=X).

Search tracking note content with optional filters

Search tracked types by name or description with relevance ranking. Multi-word input is split into words; a row matches if ANY word prefix-matches ("disk space free" finds get_free_disk_space). Rows matching more words rank first. Punctuation is ignored, so no FTS syntax is needed or honored. Common stopwords (the, of, and, ...) are ignored unless the query is nothing but stopwords. Results are capped by limit; total_count reports all matches.

Search note content in user_directives.db with optional additional filters

Search user directives with optional filters. At least one filter required.

Add a custom return statement for a helper function. Extends core return_statements with user-defined guidance, context, or preferences. To modify: delete first, then set new statement.

(For use with InterCommAIMFP) Cheap, low-token counts summary of ONE branch's changeset — add/modify/delete per entity kind, add/remove per reference kind, touched files/modules, and warning count — WITHOUT returning the full ~8k-token object. Affordable to read for every pending branch before deciding apply order. Identify the changeset by changeset_id (loads the persisted artifact) or by branch+base_commit (re-exports). Pure read.

Swap order_index of two completion paths

Update project.last_known_git_hash with current Git HEAD

Enable/disable tracking feature

Remove file links from a task, subtask, or sidequest — e.g. a file auto-linked to the wrong in_progress item. Does not touch the files.

Update completion path

Update existing directive preference in directive_preferences table

Update file metadata

Update flow metadata

Update function metadata

Move function to different file (rarely used)

Update multiple functions in a single file

Update interaction metadata

Update a single item's status, name, or description. Primary use: marking items in_progress or completed as work progresses.

Batch-update multiple items with the same field values. Primary use: marking multiple items completed at once when parent completes.

Update milestone status, name, or description. When completing, check if all milestones in the path are done to cascade completion.

Update module metadata. Only non-None fields are updated.

Update note metadata

Update project metadata

Update existing entry

Update project root in infrastructure table. Failsafe: Inserts if not exists (should never happen, initialized by SQL).

Orchestrated state change for tasks, milestones, and paths. Maps actions (start_task, complete_task, block_task, etc.) to correct status updates and note creation.

Update entry in user settings database

Update sidequest metadata

Update existing source directory in infrastructure table

Update subtask metadata

Update task metadata

Update theme metadata

Update type metadata

Update relationship role only

Update entry in user_directives.db table. Every key must be a column of the table (compared without case). On user_directives the approval columns (approved, approved_at, approved_by, approval_hash) and status='active' are refused: approval is a person's decision, recorded by the user's own path (`aimfp directives approve`, or a host calling aimfp.hooks.approval), and activation goes through activate_user_directive.

Update existing project-wide user setting in user_settings table

Returns list of allowed values for CHECK constraint enum fields in user_directives.db. Parses schema to extract CHECK(field IN (...)) constraints.

Returns list of allowed values for CHECK constraint enum fields in user_preferences.db. Parses schema to extract CHECK(field IN (...)) constraints.

Checks an action_config against AIMFP's envelope for its action_type BEFORE the row is written; add_user_custom_entry and update_user_custom_entry refuse a non-conforming config. AIMFP validates the ENVELOPE ONLY. api_call, script_execution and notification are generic, so AIMFP can specify them fully. function_call and command name a target that exists in the caller's own project, so AIMFP checks that a target is named and well-formed but never that it resolves - caller_resolved=true marks exactly that, and means 'the shape is right, now check it resolves'. Unknown keys are errors, not ignored. On failure, errors names every offending key and grammar carries the expected envelope.

Checks a trigger_config against AIMFP's grammar for its trigger_type BEFORE the row is written. add_user_custom_entry and update_user_custom_entry refuse a non-conforming config, so calling this first turns a rejected insert into a fixed config. Time triggers are a discriminated union on 'kind' (interval/daily/weekly/monthly); event, condition and manual have fixed shapes. Unknown keys are errors, not ignored - a typo'd schedule key is the silent failure this exists to surface, because a schedule that never fires produces no log line and no recorded error. On failure, errors names every offending key and grammar carries the expected shape for that trigger_type, so the AI is told what IS expected rather than guessing.

(For use with InterCommAIMFP) Pre-fan-out preflight on current main: cheap insurance against silent fan-out corruption. Asserts (a) every relevant row has a stable identity key (hierarchy slugs + function/type entity_key — i.e. backfill_semantic_keys has nothing left to do), (b) .aimfp-project/project.db is committed (clean working tree for that path) so every clone shares the same base, and (c) no schema migration is pending. Returns {ready, blockers}. Run on main BEFORE spawning workers. Pure read.