agentleFS
haqalizhaqaliz/assay/.claude/skills/assay-next/SKILL.md

assay-next

Skill0 starsChanged 3 days agoApache-2.0View on GitHub

Use when deciding what to build next in Assay and you want the single highest-leverage capability picked from the repo's own roadmap and planning files (not invented), grounded in the moat and in what has already shipped or been deferred, ending with a ready-to-run handoff. Triggers on "assay-next", "an", "what's next", "next feature", "pick next".

The file

We split this skill into the rules it gives an agent

9 rules. About 0 are common, 7 are distinctive, and 2 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.

  • 2

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Read the repo's own roadmap and planning files, rank the real candidate capabilities against the moat and against what has shipped or been …

    line 11
  2. Distinctive:

    After a merged capability or a phase gate, when choosing the next unit of work.

    line 22
  3. Very specific:

    Not for: executing a chosen capability (use assay-begin-fast), or planning an already-chosen one (use prd-interview / tech-plan).

    line 23
  4. Very specific:

    Read these (the source of truth, in this order).

    line 28
  5. Distinctive:

    If a file above is missing, still reference it by path and say it isn't written yet.

    line 45
  6. Distinctive:

    Never substitute memory for a file you couldn't read.

    line 46
  7. Distinctive:

    The gate beats the feature: Before Phase 0's gate (one real skill run end to end, with an observed behavior that differs from a declared …

    line 58
  8. Distinctive:

    Read the files above. Build the candidate list: unshipped capabilities whose dependencies are met, and follow-on slices of shipped ones.

    line 66
  9. Distinctive:

    If the strongest-looking candidate has a real blocker, say so.

    line 87

≈ “Common” is an estimate: at least 1 in 5 of the 372,429 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

Assay Next (pick the most important capability)

Overview

Read the repo's own roadmap and planning files, rank the real candidate capabilities against the moat and against what has shipped or been deferred, and recommend the single highest-leverage one to build next. End with a ready-to-paste assay-begin-fast invocation so the next session can start that worktree.

This skill RECOMMENDS and hands off. It does NOT create a worktree or start assay-begin-fast itself; the user runs the handoff prompt when ready.

When to use

  • "what should I build next", "pick the next capability", at the start of a session.
  • After a merged capability or a phase gate, when choosing the next unit of work.
  • Not for: executing a chosen capability (use assay-begin-fast), or planning an already-chosen one (use prd-interview / tech-plan).

The candidate set is the FILES, never invented

Read these (the source of truth, in this order). Assay is greenfield, so the planning docs carry almost all the signal today — but code, once it exists, wins over prose:

  • docs/technical/CAPABILITY_ROADMAP.md: the A1..A10 engine capabilities in build order, each with what it does, its dependencies, and its guardrail tie-in. This is the primary candidate set.
  • docs/ROADMAP.md: the phases (0–3), the milestones, and the R1..R9 risk register. A candidate that retires a High-likelihood/High-impact risk is worth more than its slot suggests.
  • CLAUDE.md and VISION.md: the wedge, the moat, and the guardrails the pick must obey.
  • docs/technical/ARCHITECTURE.md: the ingest → sandbox → capture → diff → verdict → bundle design.
  • docs/planning/*/: in-flight, completed, and deferred work, once any exists.
  • git log / git tag / the test suite: what actually shipped. Trust this over prose. Today there are no commits beyond the scaffold — say so plainly.

If a file above is missing, still reference it by path and say it isn't written yet. Never substitute memory for a file you couldn't read.

How to rank (grounded in CLAUDE.md)

  1. Execution-grounded verdicts only. A candidate that lets a static scan or a model's opinion stand in for an observed sandbox behavior is dropped.
  2. Respect the build order and its dependencies. A2/A3 need A1+A2; A4 needs A2+A3; A6/A7 need A5; A8 widens formats only after the core is trusted. The next capability is usually the lowest unshipped ID whose dependencies are met.
  3. The sandbox and the floor are the critical path. A2 (sandbox) and A5 (verdict/floor) are the moat (R4/R2). Nothing downstream is trustworthy without them.
  4. Deepen the corpus. Favor work that feeds the findings corpus (A7) wherever feasible.
  5. The gate beats the feature. Before Phase 0's gate (one real skill run end to end, with an observed behavior that differs from a declared one), the highest-leverage work is whatever produces that gate.
  6. Follow-on slices count. A shipped capability's next slice (a new finding class, a new strictness level) is a valid candidate.

Process

  1. Read the files above. Build the candidate list: unshipped capabilities whose dependencies are met, and follow-on slices of shipped ones.
  2. For each candidate, record: shipped-state (cite the file), dependency status, moat-leverage, the risk it retires (by R-id), and any known blocker.
  3. Rank by the rules above. Pick ONE, plus one or two alternates.
  4. Sanity-check the pick against the guardrails and the current phase gate.
  5. Produce the handoff (below).

Output format

  • The pick: one line naming the capability (with its A-id) and a kebab-case slug.
  • Why: 2 to 3 bullets tying it to the moat, its dependencies, and what shipped.
  • Alternates: one or two lines.
  • Known caveat: the nearest feasibility risk (name the R-id where one applies).
  • Handoff prompt (ready to paste): an abf feat <slug> line plus a 3 to 5 sentence inline brief including the caveat and the capability's acceptance tests.

Honesty rules

  • Ground every shipped / pending / deferred claim in a named file.
  • Almost nothing has shipped yet. Say "not started" rather than implying progress.
  • If the strongest-looking candidate has a real blocker, say so.

Common mistakes

MistakeFix
Inventing a capability not in the docsThe candidate set is A1..A10 in CAPABILITY_ROADMAP.md
Recommending a capability whose dependencies are unbuiltCheck the dependency line
Re-recommending shipped workCheck git log first, not the prose
Recommending a static-only verdictDrop it — execution (the sandboxed run) decides
Reaching for A9 (hosted) earlyIt depends on A5-A7; the core ships first
Starting the worktree from this skillOnly recommend and hand off; the user runs abf

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.