Install
$ agentstack add skill-saikaaco-saika-hermes-skills-wf ✓ scanned · ✓ verified, works with Claude Code, Cursor, and more.
Security review
✓ PassedNo issues found. Passed automated security review. · v0.1.0 How review works →
- ✓ Prompt-injection patterns
- ✓ Secret / credential exfiltration
- ✓ Dangerous shell & filesystem operations
- ✓ Untrusted network calls
- ✓ Known-malicious package signatures
What it can access
- ✓ Network access No
- ✓ Filesystem access No
- ✓ Shell / process execution No
- ✓ Environment & secrets No
- ✓ Dynamic code execution No
From automated source analysis of v0.1.0. “Used” means the capability is present in the source — more access means more to trust, not that it’s unsafe.
Verified badge
Passed review? Show it. Paste this badge into your README, it links to the public security report.
Reliability & compatibility
Declared compatibility
Compatibility is declared by the source manifest. End-to-end runtime verification is coming, see below.
We're building live execution health for every listing: tool-call success rate, median latency, uptime, and last-checked timestamps, measured, not self-reported. It isn't live yet, so we don't show numbers we can't stand behind.
How agent discovery & health will work →About
WF — Wayfinder Flow
Purpose
WF is the short entry point and ambiguity router for the full decision-to-work system:
facts unclear -> evidence handler / research
human choice unclear -> Bounded Grilling (`clarify`, one question)
interaction unclear -> `/prototype` + human reaction
many dependent unknowns -> Wayfinder LIGHT or FULL
route settled -> To-Spec when a contract is needed
approved spec needs decomposition -> `/to-tickets`
execution needs durability -> separately authorized Kanban
Manual invocation is /wf. Automatic invocation follows the standing user instruction: when work contains genuine doubt or something required is unclear, do not silently assume; investigate discoverable facts, then ask the human through this flow when a human-owned value remains.
Trigger gate
Invoke WF automatically before proceeding when at least one unresolved item can change the action, output, safety, or acceptance and any of these holds:
- two or more reasonable interpretations remain;
- a required value, scope boundary, destination, priority, or success condition
is missing or ambiguous;
- authority for a mutation, publication, cost, credential use, cleanup,
irreversible step, or cross-profile action is unclear;
- sources or instructions materially contradict one another;
- the applicable artifact, environment, target, owner, or backend cannot be
identified confidently;
- a human preference or trade-off is being treated as if it were a factual
lookup;
- the work would require inventing acceptance criteria, an exception, or an
unsupported assumption;
- interaction with a concrete artifact is needed to judge logic or UI.
Do not manufacture doubt. WF is not required when one interpretation is plainly supported by the current request, authoritative project convention, and observable evidence. A discoverable fact is not automatically a user question: look it up first. An obvious default supported by current evidence is not a silent assumption; state it when material and proceed under the original authority.
Authority boundary
- WF may pause the current work to investigate and ask; it does not expand the
original task's mutation or publication authority.
- Returning to an already-authorized parent task after clarification is not a
new phase and may resume automatically.
- A new specification artifact, prototype mutation, ticket publication,
Kanban board, implementation phase, deployment, or cleanup still follows its owning skill and approval boundary.
- Never use a recommendation, model confidence, or common practice as a
substitute for a required human decision.
- Never ask the human to perform discoverable evidence gathering.
- Stop immediately on a user stop signal.
Manual /wf behavior
When invoked without arguments, use the current conversation and active task. When invoked with a named source or decision, bind the flow to that object.
- Identify the current destination and the earliest unresolved ambiguity that
can affect it.
- Start at LIGHT unless FULL persistence criteria are already met.
- Run the routing procedure below.
- Ask at most one human question at a time.
- After the answer, re-evaluate the route and either resume the already-
authorized parent task or return the next bounded blocker.
Manual /wf does not itself authorize FULL artifacts, prototype code, ticket publication, Kanban, or implementation.
Routing procedure
1. Contain before acting
Before any action affected by the ambiguity:
- state the uncertain item internally;
- identify what would differ between plausible answers;
- pause only the affected mutation or output;
- preserve unrelated safe progress when it cannot bias the decision.
2. Investigate discoverable facts
Use direct tools and authoritative local/primary sources first. For a named Wayfinder EVIDENCE ticket, or when durable evidence is needed, load Wayfinder's references/evidence-handler.md and follow wayfinder-evidence-v1.
Classify the result:
- fact resolved — cite the evidence and continue;
- fact conflicted/stale/unsupported — expose the uncertainty and its effect;
- human value remains — route to Bounded Grilling;
- interaction required — route to Prototype;
- dependent fog remains — route to Wayfinder.
3. Ask one human decision
For a human-owned scope, preference, authority, trade-off, or acceptance value, invoke Bounded Grilling under WF standing authority.
Use clarify and provide:
- a concise self-contained decision context;
- the one decision that currently blocks progress;
- the relevant verified facts and constraints;
- one recommendation and its principal trade-off;
- up to four materially distinct options when options help;
- free-form
Otherautomatically supplied by the interface.
Never put selectable choices only in prose. Never batch independent questions. If the answer is ambiguous, ask only the smallest follow-up for that same branch.
4. Prototype when prose is insufficient
If a state/behavior or UI choice cannot be judged faithfully through words, return or consume PROTOTYPE_REQUIRED, then invoke /prototype under its artifact-authority, smoke-verification, evidence, human-reaction, and no-promotion gates.
A prototype ticket or standalone prototype cannot resolve from a model-written recommendation. It needs actual human reaction and one accepted decision.
5. Escalate dependent uncertainty
Use Wayfinder LIGHT when the ambiguity is bounded to the current session and no durable map is warranted. Use Wayfinder FULL only when its existing hard signals and artifact authority are satisfied, such as multi-session lifetime, three or more dependent route decisions, concurrent actors, or required durable recovery.
Automatic WF invocation does not by itself authorize FULL persistence. If a safe project root/backend or artifact authority is missing, ask before creating it.
6. Compile and decompose only when earned
- Use To-Spec after the route is settled and a requirements/acceptance contract
is useful or already authorized.
- Use
/to-ticketsonly when the user explicitly asks to decompose one named
approved source snapshot.
- Ticket publication remains a second approval bound to the exact graph and
backend.
- Plan and Kanban remain downstream, separately authorized consumers.
Multiple uncertainties
Order by dependency and risk:
- safety, authority, target, and irreversible-action uncertainty;
- destination, scope, and acceptance;
- architecture/execution class and external dependencies;
- interaction/prototype decisions;
- reversible local details.
Ask one question, wait, persist or carry the answer, then recompute. Do not present a questionnaire or silently answer later items. If several facts are independently discoverable, research them in parallel before asking the next human decision.
Handoff
When a durable handoff is useful, return:
schema: wf-route-v1
trigger: manual | automatic
parent_task:
uncertainty:
classification: fact | human-decision | prototype | dependent-fog | resolved
handler: evidence | bounded-grilling | prototype | wayfinder-light | wayfinder-full | to-spec | to-tickets
status: RESOLVED | QUESTION_PENDING | BLOCKED | ROUTED
source_or_decision_handles: []
resume_authority: existing-parent-task | none
persistence: none | pending | confirmed
Do not create this packet merely for ceremony in a tiny in-session question. Use it when another session/agent must recover the route or when an owning skill already requires persistence.
Completion
WF is complete for the current ambiguity when exactly one holds:
- evidence resolved it and the parent task can continue;
- one human question is pending through
clarify; - one accepted human answer is returned/persisted and the parent task can
resume;
- one prototype or Wayfinder route is explicitly returned;
- a named authority/evidence blocker prevents safe progress.
WF does not claim the whole parent task is complete.
Common pitfalls
- Assuming to preserve momentum. Pause the affected action and route the
ambiguity.
- Asking for facts. Investigate before questioning the human.
- Artificial uncertainty. Do not ask when the request and evidence support
one obvious interpretation.
- Question batching. One blocking human decision at a time.
- Recommendation as consent. A recommended default is not an accepted
value.
- Automatic FULL escalation. Persist only with an earned need and artifact
authority.
- Pipeline autopilot. To-Spec, To-Tickets publication, Plan, Kanban, and
implementation retain their own invocation/approval gates.
- Failure to resume. After clarification, return to the already-authorized
parent task without demanding redundant approval.
Verification checklist
- [ ] WF was manually invoked or a genuine route-affecting ambiguity existed.
- [ ] The affected action paused before an assumption became a side effect.
- [ ] Discoverable facts were investigated with authoritative evidence.
- [ ] Human values were not answered by the model.
- [ ] At most one human question was active.
- [ ] Any
clarifyprompt contained context, recommendation, and trade-off. - [ ] Prototype and FULL Wayfinder routes respected artifact authority.
- [ ] To-Tickets decomposition/publication and downstream execution remained
separately authorized.
- [ ] The parent task resumed only within its existing authority.
- [ ] No artificial question or unauthorized pipeline step was introduced.
Source & license
This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.
- Author: SaikaAco
- Source: SaikaAco/saika-hermes-skills
- License: MIT
Install and usage instructions live in the source repository linked above.
Reviews
No reviews yet, be the first.
Write a review
Versions
- v0.1.0 Imported from the upstream source.