Conversation
Add a run store as the first slice of the persisted/resumable-workflows
epic. Every execution mints a wf_<hex> run id and writes
workflow-runs/<runId>/{script.js,task.json} under a configurable store
root (default: the host agent dir): task.json is rewritten atomically
(tmp+rename) on every snapshot update and once more at terminal
(completed/failed/killed), so external readers always see complete JSON
with a fresh lastWriteAt.
The record is enriched at source with what the display snapshot lacked:
per-agent wall-clock startedAt/endedAt from the agent callbacks, real
token usage/cost/toolCalls via AgentSession.getSessionStats() read
before dispose (replacing the JSON.length/4 estimate for budget
accounting when stats exist), and each agent's full result text — the
80-char preview stays display-only. Explicit harness (pi|omp) and
dialect stamps ride in the record; the host is detected from the agent
dir path with config/env overrides.
Config precedence: createWorkflowTool({runStore}) > env
(PI_DYNAMIC_WORKFLOWS_RUN_STORE / _HARNESS) > <agentDir>/extensions/
pi-dynamic-workflows/config.json > defaults (enabled, agent dir).
Persistence is best-effort by design: a store failure never fails a
running (paid) workflow. getSessionStats/getAgentDir verified against
pi-coding-agent 0.78.0 (peer pin), 0.83.0, and the omp 17.x shim
surface; the stats read is typeof-guarded regardless.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01QrkRtZZJqWpnrzc3xe7V2x
zhangys2
added a commit
to zhangys2/pi-dynamic-workflows-michaelliv
that referenced
this pull request
Aug 26, 2026
…ixes Adopts camerondurham's upstream model resolution in place of the simpler one on this branch, and keeps the fixes that PR does not have. Michaelliv#27's resolution is strictly better: model patterns rather than exact ids only, `sonnet:low` thinking-level suffixes clamped to what the model supports, session services built once via createAgentSessionServices instead of per agent, the resolved model reported back for the progress line, and — a hole this branch had — `workflow` added to excludeTools so a subagent cannot recursively spawn workflows. Conflict resolutions: - agent(): kept our onUsage token reporting alongside their onModel. - Agent events: kept our `id` and added it to their onAgentModel, which arrived matching agents by label. Label matching is the bug this branch fixed; ids make the model land on the right progress row. - buildAgentInstructions: rejected their reintroduction of it. It narrates opts.isolation and opts.agentType, the no-op options this branch removed, and neither field exists on AgentOptions any more. - AgentRunOptions.model: both sides added it; kept one field with their doc comment, which describes the richer syntax. - resolveWorkflowModel/ModelResolverRegistry: deleted along with tests/workflow-agent.test.ts. Superseded by their resolveModelSpec, and their tests/agent-model-resolution.test.ts covers the same ground more thoroughly. Two model resolvers in one file would be worse than either. Also takes the getSessionStats guard from PR Michaelliv#31: stats are read before dispose() and behind a typeof check, since a host fork may not expose the method, in which case the caller falls back to its own estimate instead of failing the run. npm test green: biome, typecheck, build, 57 unit tests. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
HangxiangMa
pushed a commit
to HangxiangMa/pi-dynamic-workflows
that referenced
this pull request
Sep 7, 2026
…haelliv#29) Adds a user-configurable keywordTriggerWord with /workflows-trigger set/reset, and an explicit /workflows run <prompt> manual trigger that works even when keyword triggering is off. Default workflow/workflows behavior preserved. Resolves Michaelliv#31.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Implements the persistence half of the dynamic-workflows epic (#1), aligned with its vocabulary: each run lands under
workflow-runs/<runId>/asscript.js(the inline script verbatim, per #3) plustask.json— written atomically (tmp+rename) on every snapshot recompute and at terminal, so a crash mid-run leaves a valid, current record.What the record carries beyond today's snapshot:
AgentSession.getSessionStats()read in the agent'sfinallybeforedispose()(today the budget accounting estimates byJSON.length / 4); budget tracking prefers real totals and falls back to the estimate when stats are unavailable.startedAt/endedAtcaptured in the tool's agent callbacks.Store behavior: failures are swallowed by design — persistence never kills a paid run. Config precedence:
createWorkflowTool({runStore})→ env (PI_DYNAMIC_WORKFLOWS_RUN_STORE) → host config file under the agent dir → defaults (enabled,getAgentDir()root). Verified against pi 0.78/0.83 and the oh-my-pi shim surface.Two deliberate deviations from the epic's draft, open to discussion:
wf_<16 hex>rather than ulid (sortable-enough viastartedAt, and matches an existing downstream consumer contract);finishedAtrather thanendedAtat the record top level (same reason).9 new unit tests (id minting, config precedence, writer atomicity, terminal-once, end-to-end record with stats);
npm testgreen (35/35), biome + tsc clean. No behavior change when the store is disabled.🤖 Generated with Claude Code
https://claude.ai/code/session_01QrkRtZZJqWpnrzc3xe7V2x