next-skills
Skill15 starsChanged 16 days agoApache-2.0View on GitHub
The Next Skills protocol (NSP). When a participating skill finishes, surface its declared follow-on skills — each with an activation phrase and a one-line why — for the caller to pick, onboarding-style. Trigger via /next-skills, or automatically when a skill carrying the NSP opt-in block completes. Not /caller-options (routes one invocation's venue/mode) or /skills-help (the static library menu) — NSP recommends successor skills after one runs.
In this repository95 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- llms.txtllms.txt
- agent-mailboxSkill
- agent-outputSkill
- ask-gateSkill
- ask-meSkill
- audit-coupled-constantSkill
- budget-checkSkill
- build-prototypeSkill
- caller-optionsSkill
- cold-reviewSkill
- debug-taskSkill
- define-termsSkill
- defuddleSkill
- design-an-interfaceSkill
- diff-recapSkill
- directives-auditSkill
- domain-checkSkill
- double-checkSkill
- extract-signalSkill
- freshness-contractSkill
- github-syncSkill
- handoff-executeSkill
- handoffSkill
- headless-claudeSkill
- headless-codexSkill
- headless-geminiSkill
- headless-grokSkill
- headless-modelsSkill
- headless-update-modelsSkill
- headless-usageSkill
- human-drawSkill
- human-mergeSkill
- human-outputSkill
- human-rewriteSkill
- implement-taskSkill
- improve-codebase-architectureSkill
- insight-adversarialSkill
- insight-anglesSkill
- insight-assessSkill
- insight-critiqueSkill
- insight-cross-examineSkill
- insight-exploreSkill
- insight-premortemSkill
- insight-reference-classSkill
- insight-retroSkill
- insight-root-causeSkill
- insight-skill-gapSkill
- insight-test-assumptionSkill
- insight-value-of-informationSkill
- investigateSkill
- json-canvasSkill
- map-checkSkill
- map-initSkill
- map-maintainSkill
- map-rulesSkill
- map-thisSkill
- monorepo-depsSkill
- multi-agent-modeSkill
- The first 60 shown.
The file
We split this skill into the rules it gives an agent
3 rules. About 0 are common, 0 are distinctive, and 3 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.
3
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 99
Add a next-skills: list (see *Data contract* above).
- Very specific: line 118
Use n:int=99 (or any number after the skill's last real step) so NSP stays last.
- Very specific: line 123
If the next-skills skill is missing, unlinked, or unreadable, the opt-in block is a no-op — the participating skill completes normally with …
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 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.
Files
/next-skills
Next Skills (NSP) answers the question a skill leaves open: what now? When a participating skill finishes, NSP reads that skill's declared successors and surfaces them — each with an activation phrase and a short reason — for the caller to choose from. The caller (a human, or another agent) picks; nothing auto-runs.
NSP is invoked two ways: automatically, when a skill carrying the NSP opt-in block completes;
or directly, as /next-skills <skill> to preview a skill's declared successors.
Structured execution spec:
protocol.yon. It carries the three steps and the offer-only / empty-silent / loop-safe rules as validatable records; this file is the explanation. Keep the two in sync — edit one, update the other and refresh the@STAMPdate.
The protocol — three steps
- Read. For a direct preview, bind the explicitly requested skill; for a completion hook, bind the just-finished skill. An unresolved requested name never falls back to an unrelated predecessor. Take the selected skill’s
next-skills:front-matter list. Each entry has askill, aphrase(its activation phrase), and awhy(one line: why it might be useful next). If the list is absent or empty, stop silently — NSP surfaces nothing. - Surface. Present the recommendations for the caller to pick (runtime-mapped, below). Recommendations are offered, never auto-invoked.
- Activate. On the caller's pick, activate the chosen skill(s) via their activation phrase. On no pick, end quietly.
Surface step — runtime mapping
- Claude Code →
AskUserQuestion(multiSelect). One option per recommendation: the label is the skill name, the description is itswhy. The user may pick one, several, or none ("Other" / dismiss always declines). - Codex, or any runtime without a structured-question tool → present a numbered prose list,
each line
N. <phrase> — <why>, and wait for the caller's reply.
The Read and Activate steps are runtime-agnostic.
Data contract
The next-skills: front-matter field on a participating skill is canonical:
next-skills:
- skill: insight-assess
phrase: "/assess"
why: "Evaluate the options you just generated and pick one"Any prose ## Next steps section in a skill is derived/explanatory — keep it for nuance the
flat list cannot carry. Recommendations are unconditional offers: a flat list cannot encode
"if verdict X". A skill whose successors are conditional (e.g. insight-assess recommends
/plan-create only if the verdict is PROCEED) keeps that nuance in its prose section; the
why field may hint the condition ("…if the plan is sound").
Loop safety
Recommendations are offered, never auto-invoked — the caller gates every hop. NSP therefore
cannot chain A → B → A autonomously; a ping-pong would require the user to choose it each time.
This is NSP's equivalent of COP's depth guard. NSP also never blocks: if a skill declares no
successors, or the next-skills skill is itself unavailable, the skill completes normally.
Lifecycle position
NSP and Caller Options (COP) hook opposite ends of a skill invocation and never conflict: COP runs before execution (it routes venue/mode), NSP runs after (it recommends successors). A skill may carry both opt-in blocks; they are independent.
The opt-in block
A skill joins NSP by adding (1) the front-matter field, (2) a prose pointer in SKILL.md, and —
for dual-doc skills — (3) a closing @STEP in protocol.yon. Per-skill front-matter is
canonical. Unlike COP's pointer (placed near the top, since COP runs before the skill), the
NSP pointer goes at the end of the skill — NSP runs on completion.
1. Front-matter field
Add a next-skills: list (see Data contract above). It is metadata — the harness does not
execute it; it is read by NSP, by audit tooling, and by humans.
2. SKILL.md prose block
Paste at the end of the SKILL.md body, as the last section:
Next skills. On completion, run the Next Skills protocol (
next-skills/SKILL.md): surface thenext-skillsrecommendations from front-matter for the caller to pick. Offer only — never auto-invoke.
3. protocol.yon @STEP (dual-doc skills only)
Insert as the last @STEP, after the skill's own steps:
@STEP rid=step:nsp | n:int=99 | op=std:ai.prompt@v1 | args=[task="Run the Next Skills protocol: surface the next-skills front-matter recommendations for the caller to pick. Offer only; never auto-invoke. Skip silently if the next-skills list is empty."] | in=[] | out=[ref:nsp-offer]Use n:int=99 (or any number after the skill's last real step) so NSP stays last. Refresh the
@STAMP date when adding it.
Graceful degradation
If the next-skills skill is missing, unlinked, or unreadable, the opt-in block is a no-op —
the participating skill completes normally with no recommendations surfaced. NSP never blocks a
skill from finishing.
Standalone use
/next-skills <skill> reads a named skill's next-skills: field and previews its declared
successors without that skill having run — useful for inspecting or smoke-testing the chain.
Boundary
Not caller-options — COP routes a single invocation's venue and
mode before it runs. Not skills-help — that is the static library
menu. NSP recommends successor skills after one completes.
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
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.

