Conversation
andheroe
force-pushed
the
rewrite-for-act-endpoint
branch
3 times, most recently
from
April 30, 2026 19:36
a99b6e9 to
1d2230b
Compare
gora050
requested changes
May 1, 2026
andheroe
added a commit
to andheroe/application-skills
that referenced
this pull request
May 1, 2026
Adopts the structure and contract defined in membranehq/core:kits/app-use/kit.md across all 3,074 per-app skills. Mirrors the integrate-anything v2 update at membranehq/agent-skills#11. Major changes adopted from the kit: - "Step 2 — Use the connection" introduces /act with four dispatch styles (--api, --code, --key, --id). Inline --api is recommended for the FIRST CALL after a fresh connection — no build step, no BUILDING state, no waiting. Saved actions become optional Step 3. - Step 1 collapsed around `connection ensure` as the primary call (it covers find + create in one go). State-machine branching documents how to handle each returned state, including the explicit reconnect-existing rule for CLIENT_ACTION_REQUIRED — never create a fresh connection to replace a disconnected one. - `clientAction.agentInstructions` is the source of truth. Explicit warning against shortcutting to "paste this URL" when agentInstructions are present. - New "Error recovery" section distinguishes 401 (Membrane auth) from disconnected external-app from failed action, with `action-run-log get --details` for diagnostics. - Standalone "Proxy requests" section dropped — proxy semantics are now folded into `act --api`. - A "Running an action from the table above" snippet is added directly after the Popular actions table so the table isn't an orphan list: shows `act --key <key> --connectionKey <slug> --input '<json>'`. Targeted at activation failures observed during 04-28 to 04-30: - somerset12846's `action create` returning empty-shell action ID - crra78xrt's action-get 404 after create - A bot account's 45-retry loop on a never-runnable action - xiaooh8888's connection-patch loop on CLIENT_ACTION_REQUIRED Per-app content preserved unchanged: frontmatter, app intro, official docs link, "## <App> Overview" resource hierarchy, "## Popular actions" table, connection URL. Frontmatter version bumped 1.0 → 2.0 across all skills. Generated by scripts/rewrite-app-skills.py in the insighter repo. Addresses review feedback on PR membranedev#5 from gora050: - Restored "either browser OR print URL" auth wording (interactive mode often unavailable in agentic runtimes) - Dropped "and base-URL routing" phrasing (security review concern) - Collapsed Step 1 around `connection ensure` (find+create in one) - Added explicit `act --key` example after the popular actions table Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
andheroe
force-pushed
the
rewrite-for-act-endpoint
branch
from
May 1, 2026 13:23
1d2230b to
7737d19
Compare
Adopts the structure and contract defined in membranehq/core:kits/app-use/kit.md across all 3,074 per-app skills. Mirrors the integrate-anything v2 update at membranehq/agent-skills#11. Major changes adopted from the kit: - "Step 2 — Use the connection" introduces /act with four dispatch styles (--api, --code, --key, --id). Inline --api is recommended for the FIRST CALL after a fresh connection — no build step, no BUILDING state, no waiting. Saved actions become optional Step 3. - Step 1 collapsed around `connection ensure` as the primary call (it covers find + create in one go). State-machine branching documents how to handle each returned state, including the explicit reconnect-existing rule for CLIENT_ACTION_REQUIRED — never create a fresh connection to replace a disconnected one. - `clientAction.agentInstructions` is the source of truth. Explicit warning against shortcutting to "paste this URL" when agentInstructions are present. - New "Error recovery" section distinguishes 401 (Membrane auth) from disconnected external-app from failed action, with `action-run-log get --details` for diagnostics. - Standalone "Proxy requests" section dropped — proxy semantics are now folded into `act --api`. - A "Running an action from the table above" snippet is added directly after the Popular actions table so the table isn't an orphan list: shows `act --key <key> --connectionKey <slug> --input '<json>'`. Targeted at activation failures observed during 04-28 to 04-30: - somerset12846's `action create` returning empty-shell action ID - crra78xrt's action-get 404 after create - A bot account's 45-retry loop on a never-runnable action - xiaooh8888's connection-patch loop on CLIENT_ACTION_REQUIRED Per-app content preserved unchanged: frontmatter, app intro, official docs link, "## <App> Overview" resource hierarchy, "## Popular actions" table, connection URL. Frontmatter version bumped 1.0 → 2.0 across all skills. Generated by scripts/rewrite-app-skills.py in the insighter repo. Addresses review feedback on PR membranedev#5 from gora050: - Restored "either browser OR print URL" auth wording (interactive mode often unavailable in agentic runtimes) - Dropped "and base-URL routing" phrasing (security review concern) - Collapsed Step 1 around `connection ensure` (find+create in one) - Added explicit `act --key` example after the popular actions table Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
andheroe
force-pushed
the
rewrite-for-act-endpoint
branch
from
May 1, 2026 13:25
7737d19 to
844a3b8
Compare
Author
|
All four review comments addressed in 844a3b8:
Ready for re-review. |
gora050
requested changes
May 4, 2026
- Remove getmembrane.com link from intro (not needed in description). - Remove --tenant explainer paragraph (referenced --workspaceKey / --tenantKey, which are not introduced anywhere else in the skill). - Drop "or the Popular actions table below" hint from act --api — the Popular actions are saved Membrane actions, not raw API specs. - Remove the inline `act --code` section and the --code dispatch row. The agent does logic itself; surfacing remote-code-execution in per-app skills also raises a security-review concern. - Remove the duplicate "Running an action from the table above" subsection — already covered by step 2 (Reusable action by key). - Renumber: 2a (api), 2b (key), 2c (discover); was 2a/2b/2c/2d. Generated by scripts/rewrite-app-skills.py. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Author
|
All 6 review comments addressed in ac6a91e (template change, regenerated all 3,074 skills):
Sections renumbered 2a/2b/2c. Ready for re-review. |
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
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.
Mirror of membranehq/agent-skills#11 (already merged) applied across all per-app skills.
Why
This week several real users got stuck on the older
action create → wait → runflow:action createreturned an empty-shell action ID; everyaction rungot"Action type undefined is not supported"(EXP-4952)action get404'd on the just-created ID, agent gave upCLIENT_ACTION_REQUIRED; agent created a fresh connection instead of reconnecting and looped throughconnection patchThe new
/actendpoint accepts four dispatch styles —id,key,api,code— and the inlineapi/codepaths require no pre-built action. Agents can go from "user asks → real response" in one round-trip with noBUILDINGstate to wait through.That eliminates the entire failure mode for any agent that calls
act --apifor one-off requests. Saved actions become a power-user feature, not a prerequisite for the first call.What changed in each per-app SKILL.md
Preserved per-app content:
version: "1.0"→"2.0")## <App> Overviewresource hierarchy## Popular actionstableconnection ensurelineReplaced universal content with the v2 act-based flow (mirrors
integrate-anythingv2):act --api/act --code/act --key/act --iddispatch table.act --apiis recommended for the first call after a fresh connection — fastest path to a real response, no build/wait dance.READY/BUILDING/CLIENT_ACTION_REQUIRED/CONFIGURATION_ERROR/SETUP_FAILED)clientAction.agentInstructionsis the source of truth — explicit warning against shortcutting to "paste this URL"action-run-log get --detailsfor diagnostics)act --api)Failure-mode coverage
action createwrong-ID bug (EXP-4952)act --apiskips action create entirelyaction get404 after createact --apiskips action get entirelyact --api+ clearer state machineCLIENT_ACTION_REQUIREDScope
After this merges
The same shape is already on
agent-skills. Once this is merged, the ClawHub auto-sync (when working) should propagate the new content to the 91% of skills currently stale on ClawHub — addressing the discoverability/freshness problem we audited yesterday.Test plan
act --apiexample reads as a useful first call for an agentscripts/audit-skill-sync-content.pyin insighter) to measure ClawHub sync recovery🤖 Generated with Claude Code