docs(publishers-guide): fix apps/agents doc drift against actual platform behavior - #41
Merged
Merged
Conversation
…havior
Corrections verified against the code:
- Tool discovery: tools are declared in agent.json, not discovered via
GET /_tools (which the sidecar proxy actively blocks). Rewrote the
section and the scope/tips references to match.
- Session key for nebo.agents.invoke() is app:{id}:api, not agent:{id}:api.
- Permissions: document the enforced forms — network:{domain} (with
network:* wildcard) and subagent:{agentId} — instead of the unenforced
network:outbound / subagent:invoke spellings; note token gating for
storage/agents/janus endpoints.
- manifest.json: the runtime parser reads "type", not "artifact_type";
updated examples and field table (legacy spelling noted).
- Env vars: NEBO_API_URL uses the running server's port (not 8420);
NEBO_DATA_DIR resolves under Nebo's appdata (plugins/{slug}), not
~/.nebo/appdata/agents/{id} — same fix for the sidecar.log path.
- Crash recovery: first restart is immediate; backoff gates are
20s/40s/80s/160s capped at 5min (was documented as starting at 10s).
- Removed dead reference to docs/sme/APPS.md (file does not exist).
- Fixed core-tool collision list to the actual domain tool names.
Also corrected the supervisor backoff comment, which described a 10s
first wait the code never performs.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WrKTd3yCXjE69efvvpwUd3
…pps doc Corrections verified against the code: - Memory: the inherit_user flag no longer exists — owner identity inheritance (tacit/preferences, tacit/personality, read-only) is always on. Removed the flag from the table, example, and scoping matrix, and documented the actual resolution rules including the fail-closed behavior: context_isolated with no derivable context refuses memory writes instead of leaking into the shared scope. - Context isolation: contexts derive from the embed contextId or the session's active chat; only the agent-wide tacit/ namespace is inherited into isolated contexts. - Sidecar tools: removed the claim that GET /_tools discovery exists — agent.json is the only source of tool definitions. - Input fields: note that `name` is accepted as an alias for `key`. - apps.md: dev-workflow paths now use the platform-native user/agents/ root (matching the agents doc), replacing stale ~/.nebo examples. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01WrKTd3yCXjE69efvvpwUd3
…ation Traced every runtime claim in the agent guide against the code. Delegation (most publisher-visible errors): - The tool call form was wrong. There is no `agents` tool — delegation is `agent(resource: "registry", action: "delegate", ...)`. The old form survives only in an orphaned strap prompt file that is never included. - "There is no built-in depth limit" is false: delegation chains are capped at 2 levels (MAX_SUBAGENT_DEPTH) and the third is refused. Distinguished this from the separate channel handoff cap. - Auto-activation is in-memory only (no is_enabled flip, no trigger registration), and drops the target's soul/rules — documented both. - Scoped the "returns a text result" claim to wait: true. Removed the Followup Suggestions section entirely — the feature does not exist: no generator, no frontend chips, and no such hub broadcast event. Auto-install cascade: rewrote to match deps.rs. Install order is not plugins-first on all paths, non-marketplace skill refs are silently skipped, optional deps are skipped, failures are partial with no rollback (the agent stays installed and may auto-activate), and triggers/persona are not cascade steps. Filesystem watcher: documented that manifest.json edits are not detected, that sub-1s successive saves are discarded rather than deferred, and that edits inside a symlinked directory do not reliably propagate (the same overstated symlink claim is corrected in apps.md). Validation rules: split into hard errors vs silently-skipped bindings — an invalid workflow binding is dropped with a log warning while the agent loads normally, which the doc previously described as a load failure. Also documented the memory.topics feature (implemented, validated, and entirely absent from the guide), requires.interfaces, folder-path input substitution, string-only input defaults, and loader identity behavior; marked `defaults` as parsed-but-not-yet-enforced. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01WrKTd3yCXjE69efvvpwUd3
Contributor
Author
|
Reviewed with claims verified against code. Everything checks out (backoff seeding, depth cap = 2, NEBO_DATA_DIR per-slug, no /_tools, memory/isolation section, port 27895) except one factual regression: apps.md — tool-name collision list: changed FROM the correct current names ( One-line fix, then LGTM. (Hold merge — release coordination in progress.) |
agent (bot_tool.rs:2024), os (os_tool.rs:325), event, web, skill, message, loop, work — verified against each DynTool name() impl. The previous edit had reverted to pre-rename names. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_015oqdbhhgyktgKu9iqKb3V2
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.
Corrects the public publishers-guide where it had drifted from the shipped platform: delegation semantics, install cascade, watcher behavior, validation rules, agent.json memory/tools fields, and dev paths in the apps doc.
Docs-only — no runtime changes.
🤖 Generated with Claude Code
https://claude.ai/code/session_015oqdbhhgyktgKu9iqKb3V2