agentleFS
oeveryoevery/rsp/skills/rsp/SKILL.md

rsp

Skill1 starsChanged 8 days agoMITView on GitHub

Coordinate authorized direct, Change or Group work; select capabilities and accept results while executors load their methods. Retain recovery and evidence and prepare delivery without treating completion as closure.

The file

We split this skill into the rules it gives an agent

1 rule. About 0 are common, 1 is distinctive, and 0 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈1

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 0

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Follow explicit response language, personal instructions, then conversation language.

    line 79

≈ “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

RSP Core

Carry the authorized request through execution, applicable evidence and needed knowledge updates. Core owns the goal, boundary, capability selection, cross-capability decisions and result acceptance. Capabilities own their methods and may be invoked directly. Skills are the only workflow. A missing or unusable required Skill stops dependent work; report the incomplete installation without reading retired Rules or installing implicitly.

Select the work and responsibility

Before mutation, read nearest instructions and relevant CONTEXT; establish intent, authority and the existing owner from evidence. Use known paths and discover missing context as needed. An explicit WorkRef wins; multiple focus markers alone do not block work. Preserve dirty work and its owner. Before switching owners, check overlapping product and knowledge paths; overlap transfers no authority.

  • Direct: a local, explicit modification with a clear result and verification boundary. Interface changes, multiple files, strict checks or repeated execution alone require no Change or tracking preflight.
  • Change: one delivery/acceptance boundary whose implementation needs a shared record of related behavior, responsibility, compatibility or acceptance decisions. Actual staged delivery, handoff or deferred obligations can also require this record. It need not span sessions or split into independently deliverable parts. Read its six sections and relevant Specs; Focus is only a sparse recovery pointer.
  • Group: independently acceptable/deliverable child Changes with dependencies or integration obligations. Read the Brief then select a child; the Brief is not executable work. File count, complexity and agent count alone do not create a Group.

Read-only advice, diagnosis and review create no implementation Change. Establish actual impact before selecting new work; reconsider when authorized scope materially expands, not for routine corrections or failed checks. Continue a matching existing owner only for the same outcome or remaining obligations; shared files or modules alone establish no ownership. A short plan or design discussion alone requires no Change. Preserve explicit user and project tracking requirements.

Load the work model when tracked identities are missing from context; required tracking cannot be bypassed by relabelling work Direct. Use rsp status for needed owner, dependency or recovery facts, and JSON for exact dependency data. Required dependency checks and fresh archive readiness still apply.

Use the CLI invocation selected by the project AGENTS entry; rsp in these Skills denotes that selection. Resolve needed settings through the selected CLI's successful rsp config --json summary. Reuse it until relevant drift or recovery; raw YAML is diagnostic only. Projection failure blocks choices depending on unknown settings, while independent authorized work continues.

Select capabilities from their descriptions and the task contract. Executors reuse applicable entrypoint and method text already in context, load missing or changed instructions, and inspect sources; Core consults method details when needed for its own work or a concrete coordination question. Select the responsibility needed now:

  • rsp-shape: bounded read-only design or authorized planning; reopen planning only for material unresolved choices.
  • rsp-implement: diagnosis, authorized implementation or accepted-finding correction, including ordinary own checks.
  • rsp-doc: substantive document/Skill writing, needed knowledge writeback and final Change reconciliation.
  • rsp-verify: a requested declared-evidence result or a separately required verification obligation; ordinary implementation checks stay with Implement.
  • rsp-review: fixed-scope read-only judgment or explicitly bounded structural discovery. Supporting checks remain within that review.
  • rsp-commit: separately authorized exact local delivery; required for Git execution, with no Core substitute if unavailable.

Continue authorized same-scope work without another continuation request. A phase, role or method change alone requires no tool call, worker, review or test run. Required review, independent execution and protected operations still apply. For ownerless release checks, establish the exact candidate, check contract and effects authority, plus a baseline for any comparison; invent no Change or Verify owner.

Execute and reconcile evidence

Implement owns ordinary self-checks. Main-agent inspection can assess results, but an implementation participant cannot certify required independent review. Select extra Verify/Review only when requested, required or supported by concrete risk, not per task or per worker automatically.

Reuse traceable observations, checks and reviews while their target, acceptance boundary and relevant inputs still apply. Changed code, dependencies, configuration, tools or environment invalidate affected evidence; role changes and delivery-scope confirmation alone do not. Include excluded worktree inputs when assessing a staged slice, and recognize behavior changes in normative Specs or executable Skills even when stored as Markdown.

Check paths, diff, outcomes, omissions and evidence applicability. Reuse executor checks when applicable. Independent semantic review may use implementer test evidence; required independent execution must run. Preserve original evidence and verdicts with their existing owner; apply the current work record rules to Change summaries. Add no evidence ledger.

For required correction/re-review, load review convergence. Batch compatible findings into one correction, run covering checks, then review finding resolution and new risks in the correction. Stop for changed scope/authority, unsafe replay, missing required capability, non-convergence or no new discriminating evidence. Required failures remain incomplete.

Dependent writes require successful prerequisites; observe each required result rather than relying on a batch's final exit status. Recover tool failures within unchanged authority, inspecting uncertain effects before safe replay. Named checks and Commit's one-attempt rule remain binding.

Coordinate only when needed and allowed

Work shape does not choose agent topology. Automatic delegation defaults off. An explicit worker request or binding independent-worker requirement permits that bounded dispatch; otherwise resolve workflow.delegation before automatic delegation. When auto is enabled, the user or main agent chooses zero, one or multiple execution workers. RSP has no worker-count tiers or maximum; host constraints still apply.

Load worker coordination for actual dispatch, dependencies/shared-resource arbitration or required independent workers. Keep tightly coupled work local; never simulate missing worker identity or independence. Workers do not delegate further by default. Main retains integration and result acceptance; child passes do not establish integration coverage.

Load interruption recovery for pause/resume or unsettled host effects, and focus continuation for a recovery snapshot. Cross-session continuation refreshes current authority, owner, applicable status, Git index, dirty paths, evidence applicability and replay safety. Cancelling a wait does not stop workers or release resources.

Maintain records and prepare delivery

For tracked work, apply the current work record rules during execution. Load them only when absent from context. Update affected Tasks/Verify directly when scope, contract, design and knowledge remain accurate; do not append attempts or postpone current-state corrections until archive. Use the same work model for Blockers syntax. Passing checks or recovery alone adds no whole-document rewrite. Focus supplies only missing recovery context, never acceptance or authority.

Ordinary completion and checkpoint commits leave Change open. Before a tracked commit, inspect actual remaining Tasks: incomplete work selects a checkpoint; all Tasks complete selects delivery closeout, even with missing or failed verification. Reuse an explicit applicable checkpoint or closure choice without asking again. Also load closeout for established final delivery or an explicit close request. workflow.closeout is ask (default), auto or manual; no setting grants Git, push, publication, deployment, approval or human acceptance authority.

For necessary stable-knowledge updates, load durable writeback to establish facts, update their owners and reconcile affected records. Final delivery or closure additionally requires whole-Change convergence and consistency checks. Reuse applicable evidence and fill missing or invalidated coverage and separately required review. Already-accurate records need no edit; current readiness is a lifecycle observation, not semantic acceptance.

Operate command-owned state through RSP commands. Persist only open and archived; derive readiness and routing. A complete Task list never closes work. Reopen is exceptional recovery requiring explicit authority: load reopen recovery only when actually needed.

Other conditional procedures:

  • Architecture knowledge when coordinating work on modeled structure or responsibilities.
  • Setup and repair before initialization, managed-asset inspection, upgrade or repair.
  • Groups and dependencies for grouped work.
  • Context migration when a root/relevant legacy CONTEXT-MAP.md is found or migration requested; preserve context, discovery alone grants no migration.
  • Conflict handling for an intersecting Git operation.

Return the result

Follow explicit response language, personal instructions, then conversation language. Existing artifacts retain their language; new artifacts follow explicit instruction, effective project settings, scoped instructions, then conversation language. Keep identifiers and machine values exact; load language details only for contested choices.

Report the outcome, decisive evidence, material limits and needed next action. Distinguish observed effects, checks, review and acceptance; unverified claims stay unverified. Missing Required evidence blocks only dependent completion. No mandatory phase receipts or repeated logs. Load control outcome only for a consumer needing its technical rendering. For unfinished work retain only the owner, authority, evidence, blocker and safe continuation needed by the next actor.

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.