agentleFS
AxectAxect/skills/research-board/SKILL.md

research-board

Skill4 starsChanged 20 days agoMITView on GitHub

Hypothesis- and experiment-centred kanban over ~/.research. Cards are research questions with experiments, headline metrics, key plots and linked omp/Orca sessions; `rboard` shows the board, a resume view per card, cross-experiment comparison with a comparability guard, HTML pages, and jumps into (or resumes) the Orca session working on an experiment. Use when the user asks what they were doing on a project, what a hypothesis or experiment status is, wants to compare experiments, attach results or plots to a card, link or reopen a session, or when a work session changed a hypothesis, experiment, result or next action. Complements research-log (why) and never replaces compass, journal or Morgen.

The file

We split this skill into the rules it gives an agent

1 rule. About 0 are common, 0 are distinctive, and 1 is 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.

  • 1

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Results are stored, not recomputed on display: Every metric carries an extract spec and the experiment lists its sources with hashes, so …

    line 62

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill 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
SKILL.md

research-board

Answers "what was I testing, what has been tried, what did it show, and where do I resume" for every registered project. The unit is a card: one research question or attempt that can be started, paused or closed independently. The schema is in references/schema.md — read it before writing or editing a card. Working procedures are in references/workflows.md.

Boundary with the other layers

LayerOwnsNot the board's job
Card (projects/{slug}/cards/Q*.md)current work state, evidence verdict, next action, experiments, headline metrics, key plots, linked sessionsreasoning
journal.mdwhy a decision was made, lessonscurrent status
compass.mddirection, sub-goals, drift anchor (M4 stays Compass-based)experiment detail
state.mdone-snapshot session pointerhistory
Morgenschedule, paper writing, non-research tasks—

Three axes are kept apart: work (board column), verdict (evidence badge), run state (observation with a date). Never derive one from another.

Tool

scripts/rboard.py (PEP 723, uv run; dependency: PyYAML). If ~/.local/bin/rboard exists it is a symlink to it. Store root defaults to ~/.research; --root/$RBOARD_ROOT selects another (used for staging).

CommandPurpose
rboard [-p slug] [board --closed]text board: columns by work state, verdict badge, last experiment, next action, session state, days since verified
rboard show <slug:Qn>resume view: claim/scope, next action and blocker, what the verdict rests on, corrections and invalidations, experiments with metrics, arm comparisons, sessions
rboard compare <refs…> [--html f]arm comparison inside an experiment; alignment across experiments only for identical name/unit/scope, otherwise NOT COMPARABLE with the reason
rboard html [-o dir]static board + card pages (default ~/.research/board/, gitignored, regenerable)
rboard serve [--port 8765]serve the board at http://127.0.0.1:8765/, rebuilt from the cards on every page load. Use this for Orca's embedded browser: it blocks link clicks inside file:// pages, so a file:// board cannot open its card pages
rboard refresh [ref] [--accept]recompute stored metric values and source hashes from extract specs; dry run unless --accept
rboard lint [-p slug]schema errors (exit 1) and state warnings
rboard link <ref> --here | --terminal H | --session P [--unlink id]link the omp session working on a card or experiment
rboard open <ref> [--pick N] [--dry-run] [--no-focus]switch to the linked session's Orca terminal, or resume it there
rboard sessionslive omp sessions, their terminal handle and card link
rboard render <slug:Qn/En/pn>run a plot recipe from the project root

<ref> is slug:Qn or slug:Qn/En (Qn alone works when -p slug is given).

Rules

  1. Draft, then approve, then write. Card changes follow the research-log write rule: present the full card delta (and the journal draft when one exists) together, name each file, wait for approval, then write under flock ~/.research/.locks/{slug}.lock. link on an explicit user request and refresh --accept after showing the dry run are the only writes the tool makes itself.
  2. Reconcile before writing. For any card built from existing records, compare journal, state, compass and result files by topic. Unresolvable disagreement becomes needs_review: true with a note. Never pick the newest file by default; each of them has been the stale one.
  3. No invented structure. Engineering and exploration are cards with that kind, not hypotheses with a made-up falsifier. Backfilled falsifiers are retrospective. Unknown stays unknown.
  4. Scope carries every condition that changes a number (split, horizon, pinned vs Z2, data, band). Two numbers align only when name, unit and full scope are identical; do not edit a scope to force an alignment.
  5. Results are stored, not recomputed on display. Every metric carries an extract spec and the experiment lists its sources with hashes, so refresh can prove them or flag them stale. At most six headline metrics per experiment; n and lo/hi always.
  6. Plots are drawn from data with scienceplot-py so the style is consistent. Pre-existing figures may be attached with style: legacy (labelled "style unverified"); redraw one only when the user asks.
  7. Sessions are linked at the end of work on an experiment: rboard link <ref> --here. open never resumes a session that another process holds open.
  8. A running run not observed for 3+ days is reported as last observed running, currently unknown; never copy an old running into a new observation.

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.