Workflow authoring and early execution expose several recoverable conditions as low-context errors. This causes users to retry commands at the wrong lifecycle point and obscures the next safe action.

A user hit dependsOn entries must be objects, not strings twice while authoring a workflow, then encountered vault setup, workflow suspension, data-read readiness, external-call timeout, and repository-path errors while building and running a workflow. Subsequent workflow runs succeeded, so this is activation friction rather than a fundamental execution blocker.

Observed behavior

Workflow authoring rejects string entries in dependsOn with dependsOn entries must be objects, not strings.
workflow approve and workflow resume report no suspended run when the workflow is running or already complete.
data get plan triage reports the artifact is missing when the producing workflow has not completed.
vault put fails before the required vault exists and fails again when a secret already exists.
A long-running external Redmine call ends as workflow run terminated before completion, without enough timeout/cancellation context.
Running a model command outside a Swamp repository includes the absolute working directory in the error.

Expected behavior

Workflow authoring should guide users to the supported object form for dependsOn, preferably by normalizing an unambiguous string shorthand or by returning a schema error with a valid YAML example and the affected step.
Workflow lifecycle commands should expose the actual run state and a next action: wait/poll while running, inspect results when complete, or resume/approve only when a matching suspended run exists.
Reading data produced by an active workflow should identify the producer/run status and suggest waiting or inspecting workflow history rather than treating it as an unexplained missing artifact.
Vault setup errors should provide the exact create command; duplicate secret writes should describe the safe update/overwrite path if supported.
Timeout/cancellation errors should name the timed-out step/method, elapsed or configured timeout, and the recovery action.
Repository-not-found messages must redact or avoid displaying personal absolute paths by default.

Acceptance criteria

dependsOn validation names the invalid step and includes the supported object shape, with a unit test for a string entry.
Approve/resume errors report the latest relevant run state and actionable next command for running, completed, and absent workflows.
Data-read failures can distinguish a missing artifact from output that is pending an active producer run.
Vault-not-found and duplicate-secret errors provide idempotent setup/update guidance.
Workflow termination reports timeout/cancellation provenance at the failed step.
Non-repository errors do not disclose a user home directory or other sensitive absolute-path components.
Add focused coverage for each changed error contract, including structured JSON output.

Scope note

The underlying workflow runs recovered on retry. Prioritize clearer validation and state-aware recovery guidance over changing workflow execution semantics.

02Bog Flow

Open

9/11/2026, 9:53:14 PM

No activity in this phase yet.

03Sludge Pulse

git: no way to set author/committer dates on commit

Published model documentation is incomplete or inaccurate across different schema layouts

Improve workflow authoring validation and recovery guidance

serve: idle peer data.query remains stale when a remote datastore honors selective pulls

Extend --server-token-file to all --server subcommands

Serve Audit: Extension sink API — design and implementation

serve data.get cannot resolve models stored in managed configuration

serve model.edit does not publish managed configuration changes to the shared datastore

swamp serve daemon (user service) doesn't handle systemd 'lingering' — claims it 'starts automatically' but only runs while logged in

swamp serve daemon (user service) doesn't handle systemd 'lingering' — claims it 'starts automatically' but only runs while logged in

deprecated extensions are more prominent than I expected :)

Trigger schedule overrides require a swamp serve restart to take effect

Serve Audit Log - Phase 5: Extension sinks, alerting, compliance

Serve Audit Log - Phase 6: Full documentation

CI auto-merge broken: removed claude-review job was the PR approver

createModelTestContext mock Logger has .warn() but real runtime has .warning()

Serve Audit Log - Phase 4: SIEM integrations, bulk export, HMAC

aws-sm vault: support profile config key to select named AWS credential profiles

docs: data.latest second argument matches name OR spec, not just spec

Per-extension white listing

agent-runner: allow using the Claude CLI's stored login instead of requiring an API key

data.latest().attributes/.content unresolved in workflow assert/guard CEL for @mgreten/cli-agent extension data

Simplify CI: remove redundant jobs, automerge only on security + attestation gates

Add put_events method to event-bus type

workflow guard: run requires ${{ }}, evaluate rejects it — no guard form satisfies both

Unify serve config credential resolution via CEL expressions

Block temporary email services on signup

repo upgrade/init hardcode absolute swamp binary path in hook commands, breaking cross-machine portability

Docs: Serve audit Phase 4 — SIEM sinks, HMAC, and export reference

managedConfig: cross-process sentinel disagreement between CLI and serve

Expose swamp help to agents proactively in the swamp skill

Windows: vault roundtrip model outputs UTF-16LE instead of UTF-8

Make verification reviews tolerate a passing review without an explicit verdict

Enrich InvocationContext telemetry with datastore type and external vault detection

Windows: CLI outputs forward-slash lowercase paths instead of native backslash paths

Audit: emit auth-category events for OAuth device flow and token operations

Add create_pipeline_schedule to @webframp/gitlab

First-class scheduled garbage collection policy per datastore namespace

Document run GC retention configuration

Serve Audit Log - Phase 3: Real-time WebSocket streaming

docs(cli): document run GC retention configuration

Audit: many model/data handlers pass wildcard resource name instead of actual model name

swamp cmds don't realise the user is sshed into the system

spreadsheets-values: append and update cannot send valueInputOption, so neither can write — query-location parameters are never populated

Flaky POSIX shutdown handler SIGINT test exits 130

Test flake: buildOtelResource env var race in parallel test execution

Windows: HOME environment variable not set — no USERPROFILE fallback

Property test flake: CEL embedded interpolation with empty expression key

Adversarial review guard does not cover src/cli/ outside src/cli/commands/

Expose configured per-method definition arguments in model get --json

Lab #2037 mitigation does not cover category-specific bundle stores (datastore-bundles, vault-bundles, report-bundles)

Implemented/shipped issues should reference the implementation more clearly

Document configured method arguments in model get JSON

writeCatalogExport takes 55-150s per workflow step — serializes entire catalog as pretty-printed JSON

Separation of duties impossible: workflow.approve is authorized against the 'run' action, so approvers can start runs and operators can self-approve

Serve Audit Log - Phase 2: WAL, chain hashing, query API, complete coverage

data query omits report artifacts returned by data list and versions

swamp issue bug isn't auto-filling swamp version, OS and shell

Upgrade/Update causes odd locking issue on swamp deno

stale lock file not being cleaned up?

Document env.* namespace for model-instance globalArguments (host-portable config)

In "swamp workflow run", "Still running... [Xs elapsed]" should be considered non-essential output

Docs: document the approve grant action and separation-of-duties patterns

Docs: Add method-level grant scoping to authorization reference

Docs: document enriched model search JSON output shape

Retire or rebuild /admin/nurture once the epic lands — it is a gate, and gates are single-use

feat(nurture): /admin/nurture — who would get what, and why

feat(nurture): the nurture_send ledger and the dry-run projector, with the cold-start seal

feat(nurture): batch cohort reads — account-grain identification that carries its evidence

feat(nurture): the flow catalog and the qualification contract (pure domain)

Serve Audit Log - Phase 1: Event pipeline and durable storage

harness-marketing-funnel posts {batch:[...]} to /ingest, which is not a batch — every run silently measures zero

feat(cli): swamp inbox — read swamp-club notifications without leaving the terminal

feat(cli): tell operatives they have swamp-club notifications waiting

feat(api): a cheap unread-notification count for API-key clients

feat(nurture): deliver to the swamp-club inbox on the same dedupe decision as email

feat(nurture): arm needs-help-after-a-failed-run (NEEDS HELP)

feat(nurture): arm model-usage-without-workflows (BUILDING PRODUCT FEATURES)

feat(nurture): arm no-activation-after-30-days (ARRIVED, NEVER RAN)

feat(nurture): arm reactivation-after-7-days (DRIFTED)