rsp-verify
Skill1 starsChanged 8 days agoMITView on GitHub
Collect and report one selected RSP WorkOwner's declared check evidence, including separately required independent verification. Read-only; fixed-scope semantic judgment belongs to Review even when it includes checks.
In this repository10 other files it gives its agents.
- AGENTS.mdAGENTS.md
- author-rsp-skillsSkill
- distill-upstreamSkill
- release-acceptanceSkill
- rsp-commitSkill
- rsp-docSkill
- rsp-implementSkill
- rsp-reviewSkill
- rsp-shapeSkill
- rspSkill
The file
We split this skill into the rules it gives an agent
7 rules. About 0 are common, 5 are distinctive, and 2 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈5
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
- Very specific: line 20
Change: use its WorkRef and its Verify section for the declared boundary.
- Very specific: line 21
Group: use its Group reference and named Integration: condition in the Brief's Completion Conditions.
- Distinctive: line 23
Stop for ambiguous owner, target, required evidence or environment, a missing required comparison baseline, or a Group without a named …
- Distinctive: line 31
Use the caller's result contract when supplied, otherwise the direct result below.
- Distinctive: line 37
Create no backup, snapshot or digest merely to identify a routine check.
- Distinctive: line 44
Run a named mandatory command/check as required.
- Distinctive: line 45
Do not repair product/workflow state, manually replace command-owned RSP operations, replay one-shot work or continue with unknown mutation …
≈ “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.
Files
RSP Verify
For tracked work, read the shared work model when its identity and ownership definitions are missing from the current context. Use logical WorkRefs in commands and reports; cite file locations separately. Direct invocation needs no prior Core execution.
Collect declared-check evidence for one existing RSP WorkOwner when requested or separately required. Return one bounded read-only result. Review may run supporting checks without invoking Verify. Honor combined requests and required independent evidence; this Skill supplies no host independence. Verify is a Discipline for evidence collection, not routing, control, lifecycle, semantic review or Git delivery.
Require a declared boundary
Require one selected WorkOwner, current target, declared check contract and execution authority. A comparison needs its comparison baseline; a current-target check does not need an old revision. An explicit reference wins; otherwise resolve one FocusSet candidate from user intent and status. Multiple markers are valid.
- Change: use its WorkRef and its
Verifysection for the declared boundary. - Group: use its Group reference and named
Integration:condition in the Brief'sCompletion Conditions; child evidence stays with each child WorkRef. An explicit request may supply an in-progress temporary boundary, which must be written back before Group closeout.
Stop for ambiguous owner, target, required evidence or environment, a missing required comparison baseline, or a Group without a named boundary for the requested verification.
Read nearest instructions, selected Change/Brief, relevant contracts, blockers and the smallest production path needed to understand the check. Inspect the diff when the check concerns a change. This package needs no installed Core Skill. Do not invent checks from generic testability or replace an unexplained failure with a guessed assertion; return that symptom for read-only diagnosis.
Preserve read-only authority
Do not edit product files, Changes, Specs, Decisions, focus markers, archives, configuration, or Git state. Do not start publication, deployment, approval, or human-acceptance actions. Running a declared local test, build, typecheck, lint, browser check, or other environment check is evidence collection only and retains the authority required by that command.
Use the caller's result contract when supplied, otherwise the direct result below. Core owns response coordination, dispatch, result validation, acceptance and closeout; Verify supplies evidence, never worker identity/isolation, review-clean or archiveReady. Identity and independence require host observations.
Collect the declared evidence
Bind evidence to the declared target. For comparison or reuse, establish the earlier target and relevant inputs. Reuse traceable passes while their tested scope and relevant content, dependencies, configuration, tools, inputs and environment still apply. Run missing, failed or invalidated named checks and any explicitly required execution, current-environment observations or independent execution. Corrections and integration need fresh evidence for affected inputs or an enlarged boundary; a phase change alone requires no rerun.
Create no backup, snapshot or digest merely to identify a routine check. Use them only for an actual recovery requirement or independent identity/integrity consumer. Exact Git delivery, trusted external digests and required recovery safeguards retain their own contracts.
Compare each check's actual targets, inputs, coverage and completion with its declaration. Retain commands, scope, sources and results; distinguish reused evidence from new execution. Invocation errors or incomplete observations alone prove no product defect; exit status alone establishes no acceptance.
For a tool-only obstacle, diagnose read-only and inspect actual effects before continuing:
- Change a permitted collection method only with unchanged goal, owner, scope, baseline, authority and required evidence, and demonstrably safe replay.
- Run a named mandatory command/check as required; substitutes supply no pass.
- Do not repair product/workflow state, manually replace command-owned RSP operations, replay one-shot work or continue with unknown mutation effects.
Return one bounded result
Return exactly one canonical result from the collected evidence, with observed failure taking precedence over unavailable coverage:
fail: at least one named required check has observed failure, even when another check is unavailable.unavailable: no named required check has observed failure, but a required tool, dependency, service, credential, or environment could not be used.pass: every named required check passed within the declared boundary.
Report each check's outcome, including missing or unexecuted checks. Aggregation changes neither coverage nor stop conditions. Failure does not imply complete coverage; unavailable evidence does not erase observed failures.
Report the WorkOwner and applicable WorkRef or Group Integration: condition, target, named checks and outcomes, decisive evidence, omissions and any changed boundary. Distinguish reused from new execution; include a comparison baseline only for a comparative claim or reused evidence. Add machine fields only for an actual consumer or explicitly required schema, preserving its exact values. Do not repeat logs. Human-facing narration follows explicit user instruction, then personal instructions, then conversation language.
pass proves only declared checks, not semantic review, writeback, archive/commit eligibility, publication, deployment, approval or human acceptance. It grants no lifecycle, Git, external or acceptance authority.
Long-running verification remains active while its boundary and stop conditions hold. Elapsed time, heartbeat, polling and progress messages do not change the result.
Stop and return
Stop the affected verification if required evidence cannot be collected or observed, the required environment is unavailable, or the declared boundary changes. Use the caller's canonical stop reason only when supplied. For direct invocation, identify the missing input or changed boundary and the required next action; do not invent a completed verification result before the declared boundary is established.
On cancellation, wait until the command or owned process is observed stopped before beginning conflicting work. Return the result, decisive evidence, omissions and next action to the caller or user. Retry only with new evidence and safe replay inside the declared boundary.
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.

