orchidautomation/felspar
Turn your company's policy documents into tested, approved Decision Packs that every agent follows, and keep them current as the policy changes.
Prepares a Felspar approver to approve and release a Decision Pack on the Felspar web page - confirms the exact proposal reviewed, shows what changes and which agents are affected, and hands the person the review link to approve and release it themselves, to staging first or straight to production. Use this skill when someone asks to approve, release, ship, publish, stage, promote, roll back or roll out a pack or policy change for their agents. Not for reviewing impact without approving (reviewing-pack-changes).
Registers a company agent with Felspar and gets it working - chooses its jobs, picks Latest or a pinned release for each, and gives the agent's builder the MCP or REST setup. Use this skill when someone wants an agent, bot, Dust, n8n, Eve, Claude, ChatGPT or a script to use their company's decisions from Felspar, or asks how to connect, pin, update, rotate or revoke an agent's access, or how a connected agent should pick its job, which case ID to send, or what to do with an escalation. Not for making decisions inside a running agent; Felspar's agent tools carry that method.
Drives a company policy, or a weak existing Felspar pack, to a clean proposed Decision Pack with evals in one sitting - reads the document and any released pack, settles hard cases with the owner first in one stress pass, maps every requirement key, declares the job and subjects, fills every answer until check_pack is clean (each refusal and its fixes is the next step), tests hard phrasings with check_readings, writes and runs an agent-shaped eval set, and proposes it with a gap report and a log of what was done and why. Use this skill when someone wants to turn a policy, playbook, rubric, SOP or Google Doc into rules their agents follow, rescue a pack that holds too much or reads badly, create or revise a job, pick up a draft, or update a pack after the policy changed, even if they don't say 'Felspar' or 'pack'. Not for writing a missing policy (preparing-policy-documents), tests alone (writing-pack-test-cases), a change's impact (reviewing-pack-changes) or approving (approving-and-releasing-packs).
Turns "we want an agent for this team" into a plan of Felspar jobs - lists the decisions in the work, marks which need a Decision Pack and which stay with the agent, splits or merges them so each pack makes one decision, names each job's case, outcomes, owner and source document, says how the agent moves between jobs, and picks the one to build first. Use this skill when someone wants an agent for a team, department or workflow, asks whether something should be one pack or several, whether packs talk to each other, or whether a policy or pack is doing too much, even if they don't say 'Felspar'. Not for writing a policy (preparing-policy-documents), drafting one pack (drafting-decision-packs), tests (writing-pack-test-cases) or registering an agent (connecting-company-agents).
Helps a policy owner write or strengthen the company's own policy document so a Felspar Decision Pack can be drafted from it - interviews the owner one decision-shaped question at a time with a recommended default, or walks an expert who can't state the rules through real cases they decided and proposes the rules those cases imply, then writes paste-ready policy text in the document's own voice, never inventing a rule. Use this skill when the policy is thin, messy, missing ("it's in people's heads", "I just know"), or a FAQ rather than rules, when drafting-decision-packs says the document is "Thin" or "Not a decision policy", or when someone asks to write or fill gaps in a policy, playbook or SOP so their agents can use it, even if they don't say 'Felspar'. Not for turning a ready document into a pack (drafting-decision-packs), test cases (writing-pack-test-cases), an edited document (triaging-source-changes), a change's impact (reviewing-pack-changes), approving or connecting agents.
Explains what a proposed change to a Felspar Decision Pack would do before anyone approves it - backtests it against the current release, explains which decisions flip or become unknown, flags meaning changes and names the agents affected. Use this skill when someone asks what a policy or pack change would do, wants a backtest or a version comparison, or is preparing an approval, even if they only ask whether a change is safe. It never approves; approval is approving-and-releasing-packs.
Works through edited policy documents after Felspar flags a connected document as changed - compares the added version with the current one, explains which quoted passages, rules and jobs the edit touches, and starts a revision or records why none is needed. Use this skill when list_sources shows a document as changed, someone says a policy or playbook document was edited, or asks what needs attention in Felspar (including "the inbox"). Not for drafting the revision itself (drafting-decision-packs) or for deciding individual cases.
Writes and maintains test cases for a Felspar Decision Pack - one per outcome, both sides of every threshold, each exception and missing evidence - preferably from real past decisions, plus reading tests that check how Felspar reads different phrasings of each answer, on made-up wording and, with the person's agreement, a small sample of their real records, and agent-shaped eval sets of realistic multi-excerpt cases run against every candidate before release. Use this skill when someone wants to add, fix or review a pack's tests or examples, turn past decisions into tests, test how real wording is read, or check that a policy's edge cases are covered, including when a backtest or a live decision looks wrong or real cases keep being held. Not for drafting the pack's rules (drafting-decision-packs) or approving it.