Changelog

0.5.0-staging.15

The strata, uprali-demo, and uprali-demo-chatgpt packages carry the same skill change.

• When a strata_export_calculation_model result has downloadUrl, develop-business-case/references/decision-review-and-learning.md now has the assistant give the link and tell the user to open it to download the file, instead of writing the file into the chat. The link expires 15 minutes after the export. Without downloadUrl, the assistant still attaches the export as a chat file (#1788).

0.5.0-staging.14

• Change the strata entry in .claude-plugin/marketplace.json from "." to "./" so Claude's marketplace sync accepts the plugin at the distribution repository root (#1752).
• Remove the $schema key from the generated .claude-plugin/plugin.json files of strata and uprali-demo so Claude no longer reports an unrecognized key. The portable plugin.json and mcp.json files keep their required $schema values (#1752).

0.5.0-staging.13

The strata, uprali-demo, and uprali-demo-chatgpt packages carry the same skill change.

• After a successful strata_export_calculation_model call, develop-business-case/references/decision-review-and-learning.md now has the assistant attach the export to the chat as a .csv or .xlsx file named with the returned fileName and tell the user to download it. It no longer sends the user to the App's Download button, which does not save a file in Claude web. When the host cannot create a file, the assistant gives CSV content as a code block, or for XLSX offers a CSV export (#1783).

0.5.0-staging.12

The strata, uprali-demo, and uprali-demo-chatgpt packages carry the same skill change.

• When no App build reached the assistant after an intake, start-business-case/references/demo-initial-case.md now calls strata_list_cases with { "limit": 25, "filter": { "kind": "title", "value": "<intake title>" } } and takes the returned case with the newest createdAt, as understand-business-case and develop-business-case do since 0.5.0-staging.11. The demo App shows the short lookup result at step 2.1 instead of the full case list (#1777).

0.5.0-staging.11

The strata, uprali-demo, and uprali-demo-chatgpt packages carry the same skill changes.

• When the user refers to the new case, the conversation has no build result, and it showed an intake, develop-business-case and understand-business-case call strata_list_cases with { "limit": 25, "filter": { "kind": "title", "value": "<intake title>" } } and take the returned case with the newest createdAt. Without an intake they keep { "limit": 25 }. The bounded case-list rule of understand-business-case now requires "limit": 25 on every call and allows filter only where a rule says so (#1772).

0.5.0-staging.10

The strata, uprali-demo, and uprali-demo-chatgpt packages carry the same skill changes.

• understand-business-case now answers "how was this calculated" questions from the derivation projection of strata_read_calculation_run. It states the returned status word first, explains consumed inputs, operation IDs, pre-rounding intermediate values, and the terminal value for match results, quotes the diagnostic code, explanation, and source key for failed terminals, and reports mismatch, unavailable, and execution_failed outcomes with no trace. After TRACE_LIMIT_EXCEEDED or OUTPUT_TOO_LARGE it states the bound outcome and stops without quoting trace content. It copies decimal and rational text, including numerators and denominators, and source identities verbatim, and never mentions values from unconsumed conditional branches (#934).
• develop-business-case now guides a Decision that relies on calculated results. Before the preview it reads the current Case overview and the selected Run with strata_read_calculation_run and projection summary, and requires status successful and the same Investment Case, Decision Cycle, and Case Revision IDs as the current Case. Otherwise it stops and asks the user to select another Run, without substituting a Run or switching to non-calculated reliance. The preview sends calculation as { "reliesOnCalculatedResults": true, "calculationRunId": selectedRunId } and states that reliance covers the whole Run. After STALE_STATE it makes no second mutation, refreshes the Case and the Run, and requires a new Run selection, preview, and confirmation when the Case head changed (#1626).
• When the workspace App's model context names a built case with its case key and investmentCaseId, develop-business-case and understand-business-case pass that investmentCaseId to strata_list_assumption_cards, strata_read_calculation_results, and strata_read_approval_path and do not call strata_list_cases. When reporting approval status, develop-business-case states Assumption Card status only from the assumptionCardStatusCounts of strata_read_approval_path or from strata_list_assumption_cards in the same turn (#1758).
• After strata_export_calculation_model in a host that shows the workspace App, develop-business-case tells the user to press the Download .xlsx or Download .csv button in the export view and does not repeat the content. It hands over the file content only when the user reports no download button or no export view appears (#1759).
• understand-business-case keeps the scenario-with-Base and lineage-summary rules in its reference files instead of the entry file. The entry file links to the scenario-with-Base rule in history-and-comparison.md and reads the lineage-summary invariant from evidence-and-restricted-data.md, which brings the entry file back under its size budget. The behavior is unchanged (#1770).

0.5.0-staging.9

The strata, uprali-demo, and uprali-demo-chatgpt packages carry the same skill changes.

• When strata_read_approval_path is available and the user asks for a case's Decision or Outcome Review without a visible ID, develop-business-case and understand-business-case resolve the case and call strata_read_approval_path with its investmentCaseId. They pass the returned decisionId to strata_read_decision and the returned outcomeReviewReference to strata_read_outcome_review. When either value is null, they state that the case has no Decision or no Outcome Review (#1733).
• The Assumption Delegation reference of develop-business-case now covers Adoption. The skill finds adoptable work with strata_list_assumption_delegations and view awaiting_adoption, reads the delegation with strata_read_assumption_delegation before every attempt, presents its adoptionPreview, and calls strata_adopt_assumption_delegation_submission once after confirmation. It omits the adopt command when the preview has a blocker. On ADOPTION_STALE or CASE_REVISION_CONFLICT it reads fresh state and asks for a new confirmation. On SUBMISSION_ADOPTED or COMMAND_REUSED it stops without a second mutation (#1238).

0.5.0-staging.8

The strata, uprali-demo, and uprali-demo-chatgpt packages carry the same skill changes.

• start-business-case now offers "Build initial business case" for every template, including general_investment when the presenter stated investment figures. For general_investment, it sends the stated figures in investmentValues in the strata_recommend_case_intake call, so the button build uses them. It then lists the figures from the result's investmentFigures and says which the presenter stated and which are recommended defaults. The skill no longer asks the presenter to confirm in chat when they stated figures. After a button build, it no longer says that every figure used its recommended default.
• When no App build reached the model context, start-business-case no longer asks whether the presenter pressed the button. It calls strata_list_cases and takes the newest case with the intake title. If that case exists, it continues with it. If no such case exists and the message confirms the intake, it builds the case in chat. Otherwise it answers the message and waits for a button press or a confirmation.
• Before they state an Assumption Card's status, develop-business-case and understand-business-case call strata_list_assumption_cards in the same turn. Before they state approval status or a step record, they call strata_read_approval_path in the same turn. This also applies after a press of an App button.
• When the user refers to the new case and the conversation has no build result, develop-business-case and understand-business-case take the newest case from strata_list_cases. When the conversation showed an intake, the case must also have the intake title. If no such case exists, develop-business-case says that no new case exists yet and offers to start one.
• When strata_compare_scenario_to_base is available and the user asks to compare a named scenario with Base, understand-business-case takes runId and option.optionRevisionId from the case's latest strata_read_calculation_results result. If there is none, it resolves the case and calls strata_read_calculation_results first. It then calls strata_compare_scenario_to_base once, with scenarioLabel set to the user's scenario name, and answers with its Pairwise Results, such as NPV and payback.
• Steps 5 and 6 of the understand-business-case Read workflow, which read the selected projection or protected record and pin follow-up reads, moved from SKILL.md to the "Read and pin exact revisions" section of references/exact-reads-history-and-lineage.md. Their rules are unchanged.

0.5.0-staging.7

The strata, uprali-demo, and uprali-demo-chatgpt packages carry the same skill changes. The workspace App's "Build initial business case", "Confirm card", "Confirm all cards", "Approve step", and "Send change request" buttons now make their change in the App. The App reports each result to the agent through its model context and sends no chat message.

• After a press of "Build initial business case", start-business-case does not call strata_build_initial_case. It reads the build's case key, investmentCaseId, and runId from the App's model context and continues with the results. When the presenter's next message arrives and no build reached the model context, it asks whether they pressed the button, unless the message already says so. If they did, it calls strata_list_cases, takes the newest case with the intake title, and reads its results with strata_read_calculation_results.
• A confirmation in chat builds from the recommendations as shown in the intake result. Edits made in the view apply only when the presenter presses the button.
• For general_investment, the button build uses the recommended default for every investment figure, and start-business-case says so after the build. When the presenter stated investment figures, the skill asks them to confirm in chat instead of pressing the button. The chat build sends those figures in investmentValues.
• "Confirm card" and "Confirm all cards" confirm the cards in the App at their listed revisions, one at a time, and stop at the first failure. develop-business-case reports the confirmed cards from the App's model context and sends no strata_confirm_assumption_card for them. When the user says they pressed one of these buttons and no confirmation reached the model context, or the press covered more cards than the context names, it calls strata_list_assumption_cards and reports each card's status.
• "Approve step" and "Send change request" record the step record in the App. develop-business-case reports the step record from the App's model context and sends no strata_record_approval_step for it. When the user says they pressed one of these buttons and no step record reached the model context, it calls strata_read_approval_path and reports the latest step records.

0.5.0-staging.6

The strata, uprali-demo, and uprali-demo-chatgpt packages carry the same skill changes.

• The "Build initial business case" button in the intake view now sends a readable message instead of the build arguments as JSON. After that message, start-business-case takes the template, title, and description from the intake result. It takes the objectives from the message's "Objectives:" list. For each metric in the message's "Success metrics:" list, it copies the metric's title, objective, and key results from the intake result and takes its target and timeframe from the message.
• The assumption-delegation reference of develop-business-case adds strata_adopt_assumption_delegation_submission to its allowed tools. With it, the agent can adopt a delegation Submission as a new or replacement unlinked Assumption.

0.5.0-staging.5

The strata, uprali-demo, and uprali-demo-chatgpt packages carry the same skill changes.

• Each skill's SKILL.md is now smaller than 8,000 bytes, the size at which Codex truncates a selected skill. Sections moved without change into new reference files: normalized-draft and recovery in start-business-case, resolve-current-state and stop-and-recover in develop-business-case, and select-projection, history-and-comparison, evidence-and-restricted-data, and recovery in understand-business-case.
• When the user asks to confirm several Assumption Cards at once, develop-business-case shows one preview of the named cards. After one approval, it sends one strata_confirm_assumption_card per named card, lists the cards again, and reports each card's status.
• When strata_recommend_case_intake and strata_build_initial_case are both available, start-business-case follows the new demo-initial-case reference. It takes the decision description from one message and asks no follow-up questions. It chooses a template, shows the recommended intake, builds the case after one confirmation, and opens the results view for the new calculation Run.
• When strata_build_initial_case is available, develop-business-case names records in its replies by case key, title, or label instead of ID. When a capturedAt or observedAt value has only a date, it sends that date at T00:00:00Z. Before it offers to submit a case for approval, it reads the case's approval path and first saves an approval path when the case has none.
• When strata_read_calculation_results is available, develop-business-case calls it for a Run whose results the user asks to see, and the App shows the results view.

0.5.0-staging.4

The strata, uprali-demo, and uprali-demo-chatgpt packages carry the same skill changes.

• develop-business-case now calculates a case when strata_create_calculation_run is available. Earlier releases never calculated. The agent states only values that a Run returned. A Decision relies on calculated results only when it names a successful Run for the same Case Revision.
• When the user asks for a Run's calculation model as a file and strata_export_calculation_model is available, develop-business-case exports it as a CSV or XLSX file.
• Add the approvals reference: save an approval path, submit a case for approval, and record approval steps. An approval path never blocks or changes a Decision.
• Add the assumption-cards reference: group Assumptions into Assumption Cards and confirm a card.
• Add the assumption-delegation reference: delegate an Assumption to an Organization member, revise the request, save and submit a Response, return a Submission for correction, cancel the request, and list or read delegations.
• Add the performance reference: designate the revenue, ROI, and payback Outcome Commitments of a decided case, record its health, and read the performance registry.
• The approvals, Assumption Cards, and performance references apply only when all of their tools are available. Otherwise the agent tells the user that this Strata connection cannot record that data.
• start-business-case offers to record success metrics after it creates a case. The user can record or change them later. develop-business-case offers success metrics as starting points for the Outcome Commitments of an approve or pilot Decision.

0.5.0-staging.3

• The Uprali demo packages (uprali-demo and uprali-demo-chatgpt) now name their MCP server uprali-demo.

0.5.0-staging.2

• Add the Uprali demo plugin (uprali-demo) to the Claude Code marketplace. It installs the same three workflow skills and connects to the Uprali demo MCP server.
• Add the Uprali demo ChatGPT local plugin package (uprali-demo-chatgpt). It carries the same three workflow skills.

0.5.0-staging.1

• Label the plugin and both marketplaces as a staging preview.
• Declare the hosted MCP resource's current WorkOS Staging environment.
• Permit approved Strata operators to install and evaluate the proprietary package from a publicly readable repository.
• Document the final repository-visibility and private security-reporting checks.

0.4.0

• Publish the Strata Agent Plugin from a dedicated private distribution repository.
• Add native Claude Code and Codex marketplace metadata over one shared skill set.
• Document the Hermes Agent compatibility-preview path and browser OAuth setup.