Pi is a minimal terminal coding harness with a small core and extension points for TypeScript extensions, Agent Skills, prompt templates, themes, packages, custom models, custom providers, SDK embedding, and RPC integration. Swamp can expose its own models and workflows, but there is no native Pi harness integration. Users must wire the two systems together manually.

Pi skill support is broader than a single project directory. Pi discovers skills from:

~/.pi/agent/skills/
~/.agents/skills/
trusted project .pi/skills/ and .agents/skills/ directories
package skills/ directories and pi.skills package.json entries
settings.json skill paths
repeated --skill CLI paths

Each skill is a self-contained directory with SKILL.md, Agent Skills frontmatter, optional scripts, references, templates, and assets. Pi loads skill metadata at startup, loads full SKILL.md content on demand, and exposes skills through /skill:name commands.

Proposed solution

Add a first-class Pi harness integration to Swamp. The integration should provide a native model or extension that can:

Discover Pi installations, projects, packages, and configured skill locations.
List discovered skills with source path, name, description, validation warnings, and load status.
Validate SKILL.md files against the Agent Skills specification before Pi loads them.
Install, update, enable, disable, and remove skills without editing settings or copying directories manually.
Manage Pi package skill entries and preserve package metadata.
Expose Swamp workflows as Pi skills, prompt templates, or extensions with generated, valid metadata.
Run Pi with explicit --skill paths and capture the resulting session or JSON/RPC events as Swamp data.
Track skill versions, source packages, file hashes, and validation results for reproducible environments.
Support global and project scopes with explicit trust boundaries.
Report conflicts when the same skill name is discovered from multiple locations.

The integration should use the Pi CLI or documented SDK/RPC interfaces instead of relying on private files or implementation details.

Alternatives considered

Manual symlinks or copied SKILL.md directories. This does not manage versions, validation, scope, or provenance.
A generic shell model. This leaves Pi-specific discovery, package metadata, trust, and session integration to every user.
A third-party Swamp extension. This can prototype the integration, but native support can define stable resource types, lifecycle behavior, and compatibility testing.

References

Please consider this a request for native Swamp support for the Pi coding harness, with custom skill management as the first capability.

02Bog Flow

Open

8/17/2026, 11:03:55 PM

No activity in this phase yet.

03Sludge Pulse

serve: triggerSource not persisted to workflow run summary

dataOutputOverrides not applied during remote execution — ephemeral resources GC'd before override takes effect

extension rm also deletes the lockfile entry, and install on an empty lockfile is a silent success — rm/install is not a round trip

vault: audit trail should record put, delete, and annotate operations

ExtensionWorkflowRepository scans manifest.yaml as a workflow, emitting 'Unknown key manifestVersion' warning

--max-run-duration only applies to detached runs, not attached runs

Cancel endpoint reports status: cancelled for runs it did not actually cancel

@swamp/git: no way to amend a commit, but push can --force

Installation with new account creation times out

pre-flight checks run on orchestrator, not on worker — breaks remote execution with filesystem-dependent checks

Docs: update cancel endpoint response statuses in REST API reference and workflows page

docs: document --remote-only flag for swamp serve

Swamp serve daemon doesn't run with --hot-reload

serve: include workflowName in HealthSnapshotSchedule

Retyping a model instance strands catalog rows no command can reclaim — prune matches on model_id, delete resolves the current type (follow-up to #1009)

knowledge-base retrieve persists query output into the state resource, with lifetime infinite and the query text in the instance name

Method-name collision between a base model type and an extending extension silently drops the whole extension file — no error, doctor reports pass

Skip 50MB monolithic index rewrite on v2 shard-first path

Share single AwsS3Client instance across lock, sync, and verifier

Double shard assembly per command lifecycle (pull + push both read all shards)

pullChanged never records commitSeq for shard-first (v2) index

localHasAllRemoteEntries() permanently blocks sidecar update on partial caches

serve: workflow.history.get cannot find CLI-created runs or look up by run ID

"TRIAL ENDED" modal shown to team plan user when toggling extension visibility

Concurrent suspended workflow runs can't be disambiguated from 'approvals'; fresh runs should supersede matching-input suspended runs

I _think_ I might like a guard in the trigger yaml

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

serve dashboard: add cluster.instances and serve.config endpoints

CEL expressions have timestamp/duration but no current-time value, so no guard can express "older than N"

Support direct type execution (@type) with --server

access token mint writes to user vault, not serve's _token-secrets control plane in HA mode

Generic OTLP endpoint appends signal path after query or fragment

Issue lifecycle start drops top-level issue comments

serve dashboard: add instanceId, triggerSource, and step progress to WorkflowRunSummary

Document worker affinity (affinity: true) for workflows and jobs

Datastore MongoDB driver leaks connections — 5960 accumulated over 8 days

workflow approvals displays the latest run's prompt for every suspended run

@swamp/git: add a read-only working-tree diff for pre-commit review

docs: document --oauth-client-name flag for swamp serve

Docs: update workflow reference to include trigger in JSON output shapes

@swamp/git branch and push methods fail data naming when branch contains '/'

Docs: update workflow-placement.md for job/workflow-level placement inheritance

feat(serve): use identifiable OAuth client_name instead of random UUID

@swamp/git does not pass AbortSignal to Git subprocesses

Global Secret Manager methods build an invalid regional parent

Document model.method() support in assert expressions

No way to store code-review results — verdicts, findings, and cost are not persisted for any repo

Redaction regex boundary bug: label_base64 secrets with an unrecognized label pass through unredacted (e.g. swamp_ API keys)

User token page (/u/<user>) shows only "swamp_..." for every API key, making leaked-key identification impossible

upstream_state treats sparse-fetch branches as having no upstream

Add native support for the Pi coding harness

Cloud Build create supplies project instead of required projectId path parameter

workflows: support labels/target/platform at job and workflow level

swamp access grant create never registers the grant — a --deny created this way silently fails open

Support credentials in datastore config (vault-resolvable)

swamp serve OOMs on repos with 2000+ model definitions

Serve telemetry attribution for collectives

Extension workflows: no way to override trigger configuration

Add a list/discover method to the generated @swamp/aws types — no way to enumerate resources in a region

Document @vault= prefix for webhook secrets in serve.yaml

swamp access grant create installs an unrelated extension from the registry

docs: document SWAMP_API_KEY headless OAuth bootstrap for swamp serve

workflow-succeeded selects the globally-latest run, so siblings block each other under parallel use

swamp serve: webhook secrets should support @vault= resolution

Misleading collective_not_trusted auto-resolve error when a local source extension transiently fails to index during workflow validate

datastore setup: symlinks not counted in bytesCopied, causing misleading '(0B)' output

vault.get() CEL error embeds raw newline when wrapped by model resolver, producing broken output

Feature: user-configurable redaction rules (.swamp/redaction.yaml)

Fatal JavaScript out of memory: Ineffective mark-compacts near heap limit

Sync every signup to Attio, and backfill the existing population (#809 follow-up)

@swamp/issue-lifecycle probes /api/health, which swamp-club stopped serving when the Fresh scaffold stubs were removed

Lab lifecycle POST rejects any payload string that starts with a dollar sign, not just operator keys

Feed: detect and gracefully handle dead (404) post links (link rot)

/feed: per-post unread indicators and mark-as-read to track reading progress

/feed: filter and search by author, title/text, and topic

Add a :micdrop: emoji / reaction shortcode (swamp-native art included)

Push down the two cheap leaderboard scans: ghost-gate dictionary and eventCounts owner filter

$identify events are never consumed — anonymous web history never binds to the account

Finish the swamp_* attribute set on Attio users, and settle the two unprefixed ones (#809 follow-up)

aria-expanded is a raw boolean in 12 places — collapsed state is unannounced to screen readers