inwrkaiinwrkai/continuity/AGENTS.md

continuity

AGENTS.md1 starsChanged 3 months agoMITView on GitHub

Guide for AI agents extracting records from chats (with optional grounding sources), discovering a workspace schema, answering questions from existing bundles, visualizing on a canvas, and running event-driven automations. Read SKILL.md first for Step 0, schema setup, the 6-step workflow, query mode, and automations.

Where files are headingNew AGENTS.md files have more rules about keeping secrets out of the code, and fewer about code style.
Of the AGENTS.md files we read, the 31,774 first seen in the last six months, against the older ones: the kind of rule that gained most share, and the kind that lost most. About files like this in general, not this one, and an estimate, since a kind is a pattern in the words.
In this repositoryOne other file it gives its agents.
- continuitySkill

The file

We split this AGENTS.md into the rules it gives an agent

2 rules. About 0 are common, 0 are distinctive, and 2 are very specific to this project.

≈0
Common · estimate
Many files like this have a rule of this kind.
≈0
Distinctive · estimate
This author's own choices. Worth a look.
2
Very specific
Tied to their project's names, files or tools. Probably skip.

Common
Distinctive
Very specific

Very specific:
Read SKILL.md first for Step 0, schema setup, the 6-step workflow, query mode, and automations.
line 4
Very specific:
Default: inwrk/ in the current workspace.
line 130

≈ “Common” is an estimate: at least 1 in 5 of the 40,246 AGENTS.md files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.

Files

AGENTS.md

Continuity Agent Skill — Navigation Guide

Structure

continuity/
  SKILL.md          # Main skill manifest — read this first
  AGENTS.md         # This navigation guide
  references/       # Detailed reference files — load on demand
  sample/           # Demo inputs only (no bundled inwrk/) — see sample/README.md
    sources/
      crm/          # CSV exports for schema setup demos
      slack/        # Chat transcript for extract
      calendar/     # ICS grounding
      email/        # Email thread grounding

Quick Start

New to continuity? Read in this order:
  0. sample/README.md                 ← optional hands-on demo (no connectors)
  1. SKILL.md                         ← Step 0, setup, 6-step workflow, query, automations
  2. references/schema-setup.md       ← discover schema from CSVs/Sheets or interview
  3. references/schema-vocabulary.md  ← naming hints (Objects, Properties, Actions, Rules)
  4. references/stages.md             ← extract → review → write/lessons (+ events)
  5. references/objects.md            ← fallback default object types
  6. references/okf-output.md         ← inwrk bundle layout (incl. schema.md, events)
  7. references/link-repair.md        ← rewrite legacy `/records/…` links
  8. references/query.md              ← answering questions from an existing bundle
  9. references/canvas.md             ← visualize the knowledge base in a live canvas
 10. references/canvas-update.md      ← apply batched CRUD patches from the canvas
 11. references/events.md             ← events.jsonl audit log
 12. references/automations.md        ← confirmed Trigger → Condition → Action engine

Reference Index

File	What it covers
`references/stages.md`	Three pipeline stages: extract draft records, self-review, write bundle + conditional lessons
`references/objects.md`	Fallback default object catalog; custom types; points to workspace schema
`references/schema-setup.md`	Setup: discover schema from tabular sources or interview; confirm; evolve
`references/schema-vocabulary.md`	Suggestion vocabulary for Objects, Properties, Actions, Rules
`references/okf-output.md`	Bundle layout, `schema.md` / snapshots, events, automations, frontmatter, runs
`references/link-repair.md`	Detect and rewrite broken OKF-style `/…` links to editor-relative paths
`references/query.md`	How to answer questions from an existing `inwrk/` bundle without re-extracting
`references/canvas.md`	Suggest and build an interactive knowledge-base canvas (explorer + graph + CRUD + feed)
`references/canvas-update.md`	Apply batched create/update/delete patches from the canvas to `inwrk/`
`references/events.md`	Structured `events.jsonl` mutation log and cascade guard
`references/automations.md`	Author/confirm/evaluate event-driven automations

Pipeline Overview

optional setup: CSVs/Sheets or interview → confirmed inwrk/schema.md
        ↓
chat (+ connectors / grounding) + lessons + schema + prior summaries + anchor_date
        ↓
   Stage 1: Extract     → draft records (+ schema misfits)
        ↓
   Stage 2: Review      → revise drafts in place (+ identity/dedup)
        ↓
   Stage 3: Write       → inwrk bundle; lessons; events.jsonl; evaluate automations
        ↓
   Chat summary (propose schema bump / draft automations only with user approval)

Query path (no new transcript): index.md + records/index.md → load matches → answer with citations. See query.md.

Link repair path (legacy /records/… links): detect → rewrite to editor-relative paths → Maintenance log entry. See link-repair.md.

Visualize path (on user request): load bundle → build live canvas per canvas.md. Suggest canvas after query/extract summaries when records exist.

Canvas apply path: validate patch → write CRUD → emit events → evaluate automations → refresh canvas. See canvas-update.md.

Automations path: user confirms rules in automations.md → after writes, match events / schedules → internal actions auto-run; external MCP actions propose-and-approve. See automations.md.

Most Common Agent Mistakes

Mistake	Correct approach
Leaving a legacy `okf/` directory in place	If `inwrk/` is missing and `okf/` exists, rename `okf/` → `inwrk/` before the run
Skipping Step 0 on a GitHub-cloned skills install	Fetch/pull when `origin` is `inwrkai/continuity` (or legacy `ops-brain`) and workspace ≠ `SKILL_DIR`
Running `git pull` while developing this repo	Skip update check when workspace root is `SKILL_DIR`
Imposing only the nine defaults when sheets are available	Offer schema setup; discover the workspace schema
Silently confirming a custom schema without review	Present a proposal; write `status: confirmed` only after the user confirms
Ignoring confirmed `schema.md` in favor of defaults	Confirmed workspace schema is authoritative
Hallucinating fields not in transcript	Only extract fields clearly present or directly inferable
Forcing misfits into wrong fields	Flag schema misfits; propose changes when patterns recur
Auto-bumping schema without approval	Propose in run/chat; bump only on user approval
Merging all items of the same object type into one record	One draft = one real-world item
Inventing `record_id` for updates	Copy `record_id` only from loaded prior records
Matching updates by title only when identity keys exist	Prefer schema dedup / identity rules
Storing relative dates ("Thursday") without anchoring	Resolve against `anchor_date` to ISO 8601; keep original phrasing in citations
Writing numeric confidence scores	Use `high` / `medium` / `low` only
Loading every `records/*.md` on every run	Read `records/index.md` first; load full files only for likely matches
Updating lessons from every self-review	Prefer user corrections; add self-review lessons only for recurring patterns
Re-running extraction to answer a read question	Use query.md
Referencing external pipeline code	Skill is standalone — agent performs all stages directly
Persisting Actions / Interfaces / Intelligence in schema	Schema: Objects, Properties, Relationships, Rules only; executable Actions live in `automations.md`
Auto-opening a canvas after every run	Suggest always; open only when the user asks — see canvas.md
Writing Cursor `.canvas.tsx` on non-Cursor agents	Detect the host canvas skill; adapt file location and imports
Rendering empty graph/table sections in a canvas	Omit sections with no data; never show placeholders
Inventing records or edges for the visualization	Only embed data present in the bundle
Writing `inwrk/` from canvas runtime without agent apply	Queue ops in canvas; agent applies per canvas-update.md
Applying canvas patches without schema validation	Validate field keys and enums; reject unknown fields
Executing external automation actions without approval	Propose email/message/calendar payloads; run MCP only after user approves
Firing draft automations	Only `status: confirmed` automations execute
Re-processing events at or before `event_cursor`	Advance cursor after evaluate; process only newer events
Letting automation cascades loop	Events with `source: automation` must not re-trigger automations
Auto-confirming automations without user approval	Same ethos as schema — confirm only when the user agrees
Skipping `events.jsonl` after a successful write	Emit events for creates/updates/deletes alongside `log.md`
Writing OKF leading-slash links (`/records/…`)	Use editor-relative paths (`records/…`, `../records/…`); repair legacy bundles per link-repair.md
Citing records in chat as `/records/…`	Use workspace-relative `inwrk/records/<slug>.md`

Output Location

Default: inwrk/ in the current workspace. User may override. See okf-output.md for the full bundle structure (including schema.md, schema/vN.md, events.jsonl, and automations.md).

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

It worked
Partly
It didn't

What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

Check your setup, free