@wichayutdew/pi-workflows 2.5.0 → 2.6.0

package/src/workflow-status/render-step-detail.ts

@@ -262,6 +262,100 @@ function attemptDisplayOrdinal(
   return omittedAttempts + index + 1;
 }
 
+function renderLiveWorkerSession(
+  theme: WorkflowStatusTheme,
+  snapshot: WorkflowStatusSnapshot,
+  detail: SelectedStepDetail,
+  width: number,
+): Array<string> {
+  if (snapshot.execution?.kind !== 'subagent') return [];
+  const attempt = detail.attempts.at(-1);
+  const events = snapshot.execution.activityLog ?? [];
+  const eventLines = events.flatMap((event) => {
+    const isToolCall = event.startsWith('call ');
+    const isResponse = event.startsWith('response: ');
+    const label = isToolCall
+      ? '● Tool call'
+      : isResponse
+        ? '● Assistant'
+        : '● Worker';
+    const value = isResponse ? event.slice('response: '.length) : event;
+    return [
+      theme.bold(theme.fg(isToolCall ? 'warning' : 'accent', label)),
+      ...wrapPlain(value, width, theme),
+      '',
+    ];
+  });
+  return [
+    '',
+    theme.bold(theme.fg('accent', 'Live Worker Session')),
+    ...keyValueLines(
+      theme,
+      'worker',
+      snapshot.execution.agent,
+      width,
+      'accent',
+    ),
+    ...keyValueLines(
+      theme,
+      'state',
+      snapshot.execution.progress,
+      width,
+      'accent',
+    ),
+    '',
+    theme.bold(theme.fg('accent', '● Input prompt')),
+    ...wrapPlain(
+      attempt
+        ? `${attempt.task}${
+            attempt.taskTruncated
+              ? `\n… [${attempt.omittedTaskChars ?? 0} prompt characters omitted from the bounded checkpoint trace]`
+              : ''
+          }`
+        : 'The worker prompt is being prepared.',
+      width,
+      theme,
+    ),
+    '',
+    theme.bold(theme.fg('accent', '● Live message chain')),
+    ...(eventLines.length > 0
+      ? eventLines
+      : [theme.fg('muted', 'Waiting for the worker to emit activity…'), '']),
+  ];
+}
+
+/** Renders the active worker's prompt and live event chain as a step subpage. */
+export function renderLiveWorkerActivity(
+  theme: WorkflowStatusTheme,
+  snapshot: WorkflowStatusSnapshot,
+  selectedIndex: number,
+  width: number,
+): Array<string> {
+  const entry = buildPathEntries(snapshot)[selectedIndex];
+  const detail = selectedStepDetail(snapshot, selectedIndex);
+  if (!entry?.isCurrent || !detail || snapshot.execution?.kind !== 'subagent') {
+    return boxed(
+      theme,
+      'Live Worker Session',
+      width,
+      [
+        theme.fg(
+          'muted',
+          'Live activity is available only for the active worker step.',
+        ),
+      ],
+      'borderAccent',
+    );
+  }
+  return boxed(
+    theme,
+    `Live Worker Session · ${entry.title} · visit ${entry.visit}`,
+    width,
+    renderLiveWorkerSession(theme, snapshot, detail, width - 4),
+    'borderAccent',
+  );
+}
+
 /** Renders one selected execution-path row and its persisted evidence. */
 export function renderStepDetail(
   theme: WorkflowStatusTheme,

package/src/workflow-status/types.ts

@@ -14,6 +14,7 @@ export type WorkflowStatusExecution =
       readonly agent: string;
       readonly requestId: string;
       readonly progress: string;
+      readonly activityLog?: ReadonlyArray<string>;
     };
 
 export type WorkflowStatusSnapshot = {

package/src/workflow-status/view.ts

@@ -17,6 +17,7 @@ import { padAnsi } from './layout.ts';
 import { renderBoard, renderEmptyBoard } from './render-board.ts';
 import { buildPathEntries } from './render-path.ts';
 import {
+  renderLiveWorkerActivity,
   renderStepDetail,
   selectedStepDetail,
   stepTranscriptCacheKey,
@@ -45,7 +46,7 @@ export type WorkflowStatusViewDependencies = {
 
 type ViewportState = {
   readonly isClosed: boolean;
-  readonly mode: 'board' | 'detail';
+  readonly mode: 'board' | 'detail' | 'live';
   readonly selectedIndex: number;
   readonly scrollOffset: number;
   readonly viewportRows: number;
@@ -196,7 +197,9 @@ export class WorkflowStatusView implements Component {
       return;
     }
     if (matchesKey(data, 'escape')) {
-      if (this.state.mode === 'detail') {
+      if (this.state.mode === 'live') {
+        this.showDetail();
+      } else if (this.state.mode === 'detail') {
         this.showBoard();
       } else {
         this.close();
@@ -206,7 +209,12 @@ export class WorkflowStatusView implements Component {
     const pageSize = Math.max(1, this.state.viewportRows - 2);
     const contentHeight = Math.max(1, this.state.viewportRows - 1);
     const halfPageSize = Math.max(1, Math.floor(contentHeight / 2));
-    if (this.state.mode === 'detail') {
+    if (this.state.mode === 'detail' || this.state.mode === 'live') {
+      if (matchesKey(data, Key.tab)) {
+        if (this.state.mode === 'detail') this.showLive();
+        else this.showDetail();
+        return;
+      }
       if (data === 'gg' || (data === 'g' && this.pendingDetailTopKey)) {
         this.pendingDetailTopKey = false;
         this.setScrollOffset(0);
@@ -220,7 +228,8 @@ export class WorkflowStatusView implements Component {
       if (data === 'G') {
         this.setScrollOffset(Number.MAX_SAFE_INTEGER);
       } else if (matchesKey(data, Key.left) || data === 'h') {
-        this.showBoard();
+        if (this.state.mode === 'live') this.showDetail();
+        else this.showBoard();
       } else if (matchesKey(data, Key.down) || data === 'j') {
         this.setScrollOffset(this.state.scrollOffset + 1);
       } else if (matchesKey(data, Key.up) || data === 'k') {
@@ -288,14 +297,21 @@ export class WorkflowStatusView implements Component {
             this.transcriptCache,
             contentWidth,
           )
-        : renderBoard(
-            this.theme,
-            snapshot,
-            contentWidth,
-            false,
-            this.statusShortcutLabel,
-            this.state.selectedIndex,
-          )
+        : this.state.mode === 'live'
+          ? renderLiveWorkerActivity(
+              this.theme,
+              snapshot,
+              this.state.selectedIndex,
+              contentWidth,
+            )
+          : renderBoard(
+              this.theme,
+              snapshot,
+              contentWidth,
+              false,
+              this.statusShortcutLabel,
+              this.state.selectedIndex,
+            )
       : renderEmptyBoard(this.theme, contentWidth);
     if (snapshot && this.state.mode === 'detail') {
       this.ensureSelectedTranscripts(snapshot);
@@ -311,8 +327,10 @@ export class WorkflowStatusView implements Component {
       this.statusShortcutLabel,
       this.theme,
       this.state.mode === 'detail'
-        ? '↑↓/jk · Ctrl+D/U half-page · gg/G top/bottom · PgUp/PgDn · ←/h/Esc'
-        : '↑/↓ or j/k select · Enter/→/l inspect · PgUp/PgDn',
+        ? '↑↓/jk · Ctrl+D/U half-page · gg/G top/bottom · PgUp/PgDn · Tab live · ←/h/Esc'
+        : this.state.mode === 'live'
+          ? '↑↓/jk · Ctrl+D/U half-page · gg/G top/bottom · PgUp/PgDn · Tab/←/h/Esc overview'
+          : '↑/↓ or j/k select · Enter/→/l inspect · PgUp/PgDn',
     );
     this.state = page.state;
     return page.lines;
@@ -376,6 +394,23 @@ export class WorkflowStatusView implements Component {
     this.tui.requestRender(true);
   }
 
+  private showDetail(): void {
+    if (this.state.mode === 'detail') return;
+    this.pendingDetailTopKey = false;
+    this.state = { ...this.state, mode: 'detail', scrollOffset: 0 };
+    this.tui.requestRender(true);
+  }
+
+  private showLive(): void {
+    const snapshot = this.getSnapshot();
+    if (!snapshot) return;
+    const entry = buildPathEntries(snapshot)[this.state.selectedIndex];
+    if (!entry?.isCurrent || snapshot.execution?.kind !== 'subagent') return;
+    this.pendingDetailTopKey = false;
+    this.state = { ...this.state, mode: 'live', scrollOffset: 0 };
+    this.tui.requestRender(true);
+  }
+
   private ensureSelectedTranscripts(
     snapshot: NonNullable<ReturnType<SnapshotProvider>>,
   ): void {

package/agents/step.md

@@ -1,32 +0,0 @@
----
-name: step
-package: pi-workflows
-description: Executes one declarative Pi Workflows step under child-side policy
-systemPromptMode: append
-inheritProjectContext: true
-inheritSkills: false
-defaultContext: fresh
-completionGuard: false
----
-
-Execute the supplied workflow step exactly as instructed.
-
-The Pi Workflows child runtime selects and enforces the step's tools, MCP
-selectors, Bash policy, extension tools, skills, outcomes, and completion
-contract. Do not broaden those permissions. Follow the supplied workflow prompt
-when choosing an outcome; outcome names have no built-in domain meaning.
-
-Each step is an isolated, non-interactive child using its configured Pi
-Subagents profile and workflow prompt. Complete delegated work through
-pi-subagents' `structured_output`; `workflow_complete_step` belongs to
-main-agent workflow steps. Never call `contact_supervisor`,
-`subagent_supervisor`, or `intercom`. The workflow prompt defines artifact
-content and format, acceptance criteria, and the meaning of every outcome.
-
-For a non-success outcome, treat `summary` as an operator handoff, not a
-diagnostic transcript: lead with a plain-language decision, list each
-independent issue with its decisive evidence and the concrete action/owner, and
-end with the safe next move. Omit policy narration, raw logs, successful-check
-or clean-state notes, and assertions that the child lacks authority; state the
-prerequisite that would unblock it. Mention a passed check only when it directly
-explains the remaining issue.

package/examples/mr-comments.workflow.yaml

@@ -1,125 +0,0 @@
-# yaml-language-server: $schema=../schemas/workflow.schema.json
-version: 1
-id: mr-comments
-command: mr-comments
-description: Inspect merge-request feedback, plan changes, implement, and verify
-start: inspect
-maxStepVisits: 5
-summaryMaxChars: 4000
-steps:
-  inspect:
-    title: Inspect merge request
-    prompt:
-      file: prompts/mr-comments/inspect.md
-    subagent:
-      agent: scout
-      context: fresh
-      timeoutMs: 600000
-      turnBudget:
-        maxTurns: 12
-        graceTurns: 2
-    permissions:
-      tools: [read, grep, bash]
-      mcp: [gitlab/get_merge_request, gitlab/list_merge_request_discussions]
-      bash:
-        mode: allow-list
-        allow:
-          - executable: git
-            argsPrefixes:
-              [[status], [diff], [log], [show], [grep], [ls-files], [rev-parse]]
-          - executable: rg
-            argsPrefix: []
-    requires:
-      tools: [read, bash, mcp]
-    transitions:
-      ready: plan
-      blocked: $pause
-  plan:
-    title: Plan the change
-    prompt:
-      file: prompts/mr-comments/plan.md
-    subagent:
-      agent: planner
-      context: fresh
-      timeoutMs: 600000
-      turnBudget:
-        maxTurns: 16
-        graceTurns: 2
-    permissions:
-      tools: [read, grep, bash]
-      mcp: [gitlab/get_merge_request, gitlab/list_merge_request_discussions]
-      skills: [superpowers:brainstorming]
-      bash:
-        mode: allow-list
-        allow:
-          - executable: git
-            argsPrefixes:
-              [[status], [diff], [log], [show], [grep], [ls-files], [rev-parse]]
-          - executable: rg
-            argsPrefix: []
-    requires:
-      tools: [read, mcp]
-      skills: [superpowers:brainstorming]
-    gate:
-      provider: plannotator
-      submitOutcome: submit
-      approvedOutcome: approved
-      rejectedOutcome: changes-requested
-      timeoutMs: 30000
-    transitions:
-      approved: implement
-      changes-requested: plan
-      blocked: $pause
-  implement:
-    title: Implement the approved plan
-    prompt:
-      file: prompts/mr-comments/implement.md
-    subagent:
-      agent: worker
-      context: fresh
-      timeoutMs: 1200000
-      artifacts: true
-    permissions:
-      tools: [read, grep, edit, write, bash]
-      skills: [superpowers:test-driven-development]
-      bash:
-        mode: allow-list
-        allow:
-          - executable: git
-            argsPrefixes: [[status], [diff]]
-          - executable: npm
-            argsPrefix: [test]
-    requires:
-      tools: [read, edit, bash]
-      skills: [superpowers:test-driven-development]
-    transitions:
-      ready: verify
-      retry: implement
-      blocked: $pause
-  verify:
-    title: Verify the result
-    prompt:
-      file: prompts/mr-comments/verify.md
-    subagent:
-      agent: reviewer
-      context: fresh
-      timeoutMs: 600000
-      turnBudget:
-        maxTurns: 16
-        graceTurns: 2
-    permissions:
-      tools: [read, grep, bash]
-      bash:
-        mode: allow-list
-        allow:
-          - executable: git
-            argsPrefixes: [[status], [diff]]
-          - executable: npm
-            argsPrefix: [test]
-    requires:
-      tools: [read, bash]
-    transitions:
-      passed: $done
-      failed: implement
-      retry: verify
-      blocked: $pause

package/examples/prompts/mr-comments/implement.md

@@ -1,17 +0,0 @@
-Implement the approved plan with a small, coherent diff.
-
-Approved review artifact:
-
-{{reviewed.artifact}}
-
-Use repository conventions and the allowed skill. Add or update focused tests
-where practical. Treat the approved plan as the final contract. Apply TDD with
-commands permitted by this step's declarative Bash allow-list. Do not perform
-remote actions.
-Complete with `ready` when implementation and focused checks are complete;
-include a compact verification handoff in the summary.
-Complete with `retry` when the approved contract remains valid and another
-bounded attempt can safely continue unfinished work. The `replan` outcome is
-unavailable. When the approved contract is missing, stale, contradictory, or
-requires a material change to its targets or authority,
-complete with `blocked` and preserve the existing worktree for follow-up.

package/examples/prompts/mr-comments/inspect.md

@@ -1,5 +0,0 @@
-Inspect the merge request described by:
-
-{{workflow.input}}
-
-Read the relevant discussion and repository state. Separate facts from assumptions. Do not modify files. When the requested work is clear, complete with `ready`. If identity, access, or requirements are unresolved, complete with `blocked`.

package/examples/prompts/mr-comments/plan.md

@@ -1,54 +0,0 @@
-Create a concrete, decision-ready implementation plan from the inspected
-feedback. The previously rejected artifact is:
-
-{{gate.artifact}}
-
-Account for previous Plannotator feedback:
-
-{{gate.feedback}}
-
-When feedback is non-empty, revise the rejected artifact and submit the complete
-proposal for another review. Each rejection returns to this same planning step.
-
-Write the artifact for a human reviewer first. Use this order:
-
-# <Short outcome-oriented title>
-
-## Review summary
-
-In three to five plain-language bullets, explain the desired result, why it
-matters, what is in scope, and what is deliberately out of scope.
-
-## Review focus
-
-List only choices that materially affect behavior, scope, risk, or an
-irreversible action. For each choice, give the recommendation, useful
-alternatives, and consequence. Say `No decisions needed` when there are none.
-
-## Proposed changes
-
-Use a short numbered list. Name the exact file or symbol, state its observable
-change and reason, and give the corresponding acceptance criterion. Use
-checkboxes only for acceptance criteria, not for scope, evidence, risks, or
-every sentence.
-
-## Validation
-
-Explain the focused and independent checks in reviewer language. State what
-each check proves; keep exact shell metadata out of this narrative.
-
-## Risks
-
-List only material risks, with a concrete safeguard or rollback signal for
-each. Cite decisive evidence inline instead of dumping raw exploration logs or
-labeling every bullet `FACT`, `HYPOTHESIS`, or `UNKNOWN`.
-
-Do not leave `UNKNOWN`, `TBD`, or implementation-time research in the plan. If
-the allowed resources cannot resolve required access, provenance, repository
-state, resource identity, or targets, complete with `blocked` instead of
-submitting an incomplete plan. A user choice may remain only in Review focus
-with evidence and an adopted recommendation.
-
-When ready, call `structured_output` with a `value` whose outcome is `submit`
-and whose `artifact` is the full Markdown plan. This format is defined by this
-workflow prompt; Pi Workflows treats the artifact as opaque.

package/examples/prompts/mr-comments/verify.md

@@ -1,9 +0,0 @@
-Review the implementation against the requested merge-request feedback and
-approved plan:
-
-{{reviewed.artifact}}
-
-Inspect the diff and run checks permitted by this step's declarative Bash
-allow-list. Complete with `passed` only when evidence supports completion.
-Complete with `failed` to return to implementation with a compact actionable
-summary, or `blocked` when the workflow or environment must be repaired.

package/examples/settings.yaml

@@ -1,27 +0,0 @@
-# yaml-language-server: $schema=../schemas/settings.schema.json
-version: 1
-allowProjectWorkflows: false
-statusShortcut: ctrl+alt+w
-permissionCeiling:
-  tools: [read, grep, bash]
-  mcp: []
-  extensions: []
-  skills: []
-  bash:
-    mode: allow-list
-    allow:
-      - executable: git
-        argsPrefixes:
-          [[status], [diff], [log], [show], [grep], [ls-files], [rev-parse]]
-      - executable: rg
-        argsPrefix: []
-  subagent:
-    agents: [scout, planner, worker, reviewer]
-    contexts: [fresh]
-    models: []
-    maxTimeoutMs: 900000
-    maxTurns: 40
-    maxGraceTurns: 3
-    maxToolCalls: 100
-    artifacts: false
-    retryToolFailures: false

package/examples/starter-kit/steps/mr-review/publish.md

@@ -1,48 +0,0 @@
-You publish only the exact Plannotator-approved hosted review. Do not change
-local repository state.
-
-Review input:
-{{workflow.input}}
-
-Approved review:
-{{reviewed.artifact}}
-
-Approval feedback:
-{{reviewed.feedback}}
-
-Previous step handoff:
-{{last.summary}}
-
-Parse the approved Publication contract. Refresh the same review and head SHA
-through configured read-only MCP/CLI/cURL calls. For every action, first query
-the public review collections and skip it only when its exact marker, body,
-head, and anchor are already observable.
-
-Execute each remaining approved action once through its exact MCP tool/input or
-standalone `glab`, `gh`, or authenticated cURL command. Require a successful
-same-host, review-correlated response and record its remote identifier or URL.
-Never alter the approved body, target another head or anchor, expose
-credentials, approve, merge, resolve, close, delete, push, cross hosts, or add
-an unlisted action.
-
-For a GitLab inline discussion, the API requires `position` as a nested object.
-Prefer an exact configured GitLab MCP mutation whose schema accepts that object.
-When using `glab api`, submit every discussion field as multipart form data with
-`--form`, including `position[base_sha]`, `position[start_sha]`,
-`position[head_sha]`, `position[position_type]`, both paths, and the applicable
-line. Do not use `--field` or `--raw-field` for any `position[...]` key: those
-flags serialize JSON scalar keys and do not construct GitLab's nested position,
-which can create an unanchored general discussion. Before calling
-`structured_output`, fetch the returned discussion and require its note to have
-the approved position/path/line as well as the approved body and marker.
-
-After a mutation-capable call is attempted, ambiguity is `blocked`; do not
-blindly replay it. Call `structured_output` alone with outcome `published` only
-after every approved effect succeeded now or was proven already present.
-Summarize the URL/head, per-action pre-state, attempted/skipped result, exact
-correlation, and remaining work. When the previous handoff is an actionable
-verification finding, treat it as a corrective publication handoff: re-check
-the exact approved effect and retry only when its absence is conclusive. Never
-use it to alter the approved content, target, or action list. Use `blocked`
-with the same ledger when freshness, execution, or correlation is ambiguous or
-unsafe.

package/examples/starter-kit/steps/mr-review/review.md

@@ -1,76 +0,0 @@
-You are the independent code-review child. Produce one complete review proposal
-before Plannotator opens. Do not modify local or remote state.
-
-Review input:
-{{workflow.input}}
-
-Fetched evidence:
-{{last.summary}}
-
-Previously rejected artifact:
-{{gate.artifact}}
-
-Feedback from a previously rejected review:
-{{gate.feedback}}
-
-When feedback is non-empty, revise the rejected artifact against current
-evidence and submit the complete review proposal for another review. Each
-rejection returns to this same review step.
-
-Treat the fetched packet as evidence, not a verdict. Refresh the same-host head
-SHA, diff, checks, and discussions with configured read-only MCP/CLI/cURL calls.
-Inspect changed code, callers, tests, repository instructions, and relevant
-history. Review correctness, regressions, security, concurrency, compatibility,
-maintainability, and missing tests. Remove false, stale, duplicate,
-stylistic-only, and non-actionable findings.
-
-This user-owned prompt defines the Plannotator artifact:
-
-1. `# Review: <short verdict>`
-2. `## Verdict`
-3. `## Findings` — severity, exact changed-line anchor, causal problem, impact,
-   evidence, and smallest useful requested change; write `No actionable
-findings.` when clean.
-4. `## Validation`
-5. `## Publication contract`
-6. `## Safety boundaries`
-
-`## Publication contract` contains one fenced `json` object with an `actions`
-array. Each action chooses one configured mechanism:
-
-```json
-{
-  "actions": [
-    {
-      "mechanism": "mcp or bash",
-      "server": "gitlab or github when mechanism is mcp",
-      "tool": "exact configured MCP tool when mechanism is mcp",
-      "input": {},
-      "command": "exact standalone glab, gh, or curl command when mechanism is bash",
-      "effect": {
-        "host": "same host as the review",
-        "reviewUrl": "canonical review URL",
-        "headSha": "reviewed head SHA",
-        "kind": "inline-comment or review-summary",
-        "path": "changed path when inline",
-        "line": 1,
-        "body": "exact public text",
-        "marker": "stable unique marker"
-      }
-    }
-  ]
-}
-```
-
-Omit fields that do not apply to the selected mechanism or effect. Include one
-inline public comment per approved finding, or one ordinary public
-non-approving summary when clean. Every remote input and exact effect must be
-fully present with no placeholders or credentials. The actions may only post
-this review on the same host and head; never approve, merge, resolve, close,
-delete, force-push, cross hosts, or perform an unlisted mutation.
-
-Call `structured_output` alone with outcome `submit`. Put the complete Markdown
-review in `artifact` and a self-contained review/publication handoff in
-`summary`. Use `blocked` when current evidence cannot support a safe,
-publishable review. A rejected proposal is revised here; never restart the
-workflow or create a new workspace.

package/examples/starter-kit/steps/mr-review/verify.md

@@ -1,35 +0,0 @@
-You are the final read-only verification child for the published hosted review.
-Do not execute a publication action or mutate local or remote state.
-
-Review input:
-{{workflow.input}}
-
-Approved review:
-{{reviewed.artifact}}
-
-Publication ledger:
-{{last.summary}}
-
-Parse the approved Publication contract and refresh the same review through
-configured read-only MCP/CLI/cURL calls. Require the current head SHA to match
-the approved artifact. Independently query the public discussion, note, or
-review collections and prove every exact marker, body, head, effect kind,
-optional path/line anchor, and remote identifier. The publication ledger alone
-is not proof.
-
-Call `structured_output` alone with:
-
-- `verified` only when all approved effects are observable exactly once or in
-  the explicitly idempotent form described by the contract;
-- `failed` for an actionable, unambiguous missing or mismatched approved
-  effect. Its self-contained summary becomes the next publication worker's
-  corrective handoff, so include the expected and observed state, exact
-  evidence, and the smallest safe repair;
-- `retry` for a transient read-only verification failure after safe equivalent
-  checks were attempted;
-- `blocked` only when the review, head, target, or remote result is stale,
-  ambiguous, or unsafe to repair automatically.
-
-For `verified` and `failed`, summarize the canonical URL, current head,
-verified or missing remote identifiers/URLs and anchors, action count, and
-final verdict. Do not mutate state yourself.