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

research-log

Skill4 starsChanged 20 days agoMITView on GitHub

Maintain a research log that acts as an advisor, not just a diary. Registers projects; records decisions with generalizable lesson extraction; saves session state; recalls cross-project lessons and prior solutions when stuck; checks proposed approaches against promoted anti-pattern rules; runs weekly review with lesson promotion and dashboard refresh. Use when the user asks to initialize a research-log project, record an experiment or decision, save session state, recall past lessons or how a blocker was solved before, check a planned approach against past mistakes, or review all projects.

1 other project uses a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

11 rules. About 0 are common, 7 are distinctive, and 4 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈7

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 4

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    M3 — recall when stuck: Search across all projects for prior lessons and solutions.

    line 16
  2. Very specific:

    Read ~/.research/rules.md (small, promoted-rules-only.

    line 25
  3. Distinctive:

    Check whether any Rule's trigger/Check matches the proposed action.

    line 26
  4. Distinctive:

    If a Rule matches, surface it first: "You are about to {X}.

    line 27
  5. Distinctive:

    If nothing matches, say nothing about rules and proceed normally.

    line 29
  6. Distinctive:

    Read the matching workflow reference for the user's request:

    line 37
  7. Very specific:

    Follow the file schemas defined in references/conventions.md exactly.

    line 54
  8. Very specific:

    When recording a decision, always attempt lesson extraction.

    line 57
  9. Distinctive:

    Treat dashboard updates as review-time derived data unless a workflow says otherwise.

    line 59
  10. Distinctive:

    When a workflow requires user approval before writing, draft first and wait for approval.

    line 60
  11. Very specific:

    Never write to a legacy ~/.research-log/ backup.

    line 61

≈ “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 Log

This skill maintains a unified research-log workspace in ~/.research/ and uses it to actively improve research, not merely to record it. The data model and all file schemas are defined in references/conventions.md — read it first whenever you touch the store.

The skill earns its keep at three moments:

  • M1 — decision-time guardrail. Surface relevant anti-patterns before the user commits to an approach. See "Decision guardrail" below — this fires from conversation, not only on command.
  • M3 — recall when stuck. Search across all projects for prior lessons and solutions.
  • M4 — real-time drift. Flag work that has drifted from the current Compass focus.

Decision guardrail (M1) — always active

Whenever the user proposes a research approach — an architecture, model, loss function, methodology, experimental design, or "let's do X" choice — before agreeing, elaborating, or helping execute it:

  1. Read ~/.research/rules.md (small, promoted-rules-only; cheap to load).
  2. Check whether any Rule's trigger/Check matches the proposed action.
  3. If a Rule matches, surface it first: "You are about to {X}; this bit you in {projects} — {Check}." Include the evidence. Let the user decide; do not block.
  4. If nothing matches, say nothing about rules and proceed normally.

This is a behavioral directive, not a hook. It applies in any registered project's working directory. Keep it lightweight — only promoted Rules are checked, and only when an actual approach choice is on the table (not for routine edits or questions).

Workflow router

Read the matching workflow reference for the user's request:

  • Project registration or onboarding → references/initialize.md
  • Recording an experiment, debugging result, or design choice → references/record.md
  • Saving current session context → references/save-state.md
  • Recalling past lessons / how a blocker was solved before (M3) → references/recall.md
  • Explicitly checking a planned approach against past mistakes (M1) → references/check.md
  • Weekly review, dashboard refresh, lesson promotion, archive scan → references/review.md
  • Current status of a hypothesis or experiment, the board, comparing experiments, attaching results or plots, or resuming the session working on an experiment → the research-board skill (cards under projects/{slug}/cards/). Decision reasoning and lessons stay here; when an entry concerns a card, add a **Card**: {slug}:Q{n}/E{m} line to the new journal entry.

Working rules

  • Match the current working directory against the Project Registry in ~/.research/dashboard.md before acting on an existing project.
  • Follow the file schemas defined in references/conventions.md exactly.
  • Acquire the per-project lock (~/.research/.locks/{slug}.lock) before writing project state.
  • Compass ← current focus is the anchor for drift detection; keep exactly one.
  • When recording a decision, always attempt lesson extraction; when a lesson recurs across ≥ 2 projects, propose promoting it to a Rule and regenerate rules.md.
  • Treat dashboard updates as review-time derived data unless a workflow says otherwise.
  • When a workflow requires user approval before writing, draft first and wait for approval.
  • Never write to a legacy ~/.research-log/ backup; it is read-only.

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.