docs: agent-compose.yaml design spec (TASK-098)#228
Merged
Conversation
tiwillia
reviewed
Mar 17, 2026
docs/design-docs/agent-compose.md
Outdated
|
|
||
| --- | ||
|
|
||
| ## Drift Detection |
There was a problem hiding this comment.
Add a Why section here - asking because I don't understand why its required.
The application is getting heavy on features, we should be able to spell out real use cases where I would want automatic drift detection and resolution in the application.
If we abandon drift detection / resolution, it simplifies this feature quite a bit, as we don't need to store any state in the application.
Owner
Author
|
Updated the spec based on @tiwillia's feedback and additional direction: server-side drift tracking is removed entirely. The revised model is terraform-style: the YAML file is the source of truth, and |
tiwillia
reviewed
Mar 17, 2026
|
|
||
| ## CLI Architecture — Import as a Client-Side Tool | ||
|
|
||
| `boss import` is implemented entirely in the CLI. It does not call a single monolithic server endpoint. Instead, it: |
There was a problem hiding this comment.
I can imagine some use cases that will require this to be supported in the server as well (like adding this capability to the MCP server) but +1 to move forward, this is simple and easy to expand upon in its current state.
Portable team blueprint format for exporting/importing agent hierarchies. Covers: YAML schema, sync import semantics, drift detection, CLI + UI surface, and 3-phase implementation plan. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
…edback Key changes: - Import repositioned as CLI tool calling existing endpoints (not a monolithic POST /spaces/import). Server exposes primitives; CLI orchestrates the import plan (terraform model). - Remove server-side drift tracking (fleet_config_hash, drift endpoints). Diff is computed client-side by comparing local YAML to fetched server state — same as kubectl apply / terraform apply. - Import UI modal uses same client-side diff approach. - Add --restart-changed flag (replaces --restart-drifted) for immediately respawning agents whose config changed after import. - Address reviewer comment (tiwillia): remove server-side state tracking, keep import/export simple with no new server concepts. - Security section retained: command allowlist, YAML bomb protection, prompt injection notes, prune+live-session safety, auth. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
P0: - Remove false "persona sandboxing" claim; replace with accurate statement that prompts are stored as-is, operators must review before importing from untrusted sources - Dry-run now shows full persona prompt diffs (not just one-line summary); malicious prompts are visible before apply P1: - shared_contracts: explicit delivery mechanism documented (prepended to ignition text before personas + initial_prompt) - Composition order rule: shared_contracts → persona(s) → initial_prompt - Persona ID namespace: global namespace documented as known tradeoff; guidance to use unique IDs (e.g. project-prefixed) - Server-side YAML bomb guard: enforced independently of CLI, covers UI upload path (1MB / 100 agents) - repos URL validation: HTTPS-only, RFC 1918 + link-local blocked, DNS pre-check before passing to ambient runner - work_dir validation: absolute path required, .. rejected after Clean, BOSS_WORK_DIR_PREFIX configurable allowed prefix - Post-import state: "N agents created, none running yet" + Spawn all action (CLI: --spawn-after-import; UI: button in success modal) - Phase ordering fixed: export + import core ship together (Phase 1); --prune and --restart-changed move to Phase 2 P2: - UI diff XSS: all YAML-sourced content rendered as plain text, no innerHTML of user-controlled strings - Export URL credential scrubbing: userinfo stripped from repo_url - agent description field added to schema table - Round-trip fidelity: export always includes all fields explicitly (even defaults) to prevent phantom diffs on re-import - Space creation edge case: offer to create if missing, or error with --no-create-space flag P3: - Frontend YAML parser dependency noted (js-yaml or equivalent) - --yes CI/CD safety note: prompt injection risk documented - Command allowlist location: BOSS_COMMAND_ALLOWLIST env var - Concurrent import: 409 treated as already-done, safe to continue - --restart-changed uses pre-apply diff to avoid concurrent-change races Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
jsell-rh
force-pushed
the
feat/agent-compose-spec-task098
branch
from
e88e1cb to
599cc8a
Compare
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
2 participants
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.
Summary
Design spec for
agent-compose.yaml— a portable team blueprint file for exporting and importing agent hierarchies between spaces/environments.Key design decisions:
kubectl applystyle): create new agents, update existing configs, leave unmentioned agents alone (opt-in--prune)fleet_config_hashper agent; show "out of sync" badge + "Sync fleet" button when config diverges from last importDesign discussion summary
Test plan
Closes TASK-098 (design phase)
🤖 Generated with Claude Code