@archme/cli 0.25.0 → 0.25.1
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/dist/cli.js +4 -1
- package/package.json +1 -1
- package/sandbox-image.json +1 -1
package/dist/cli.js
CHANGED
|
@@ -11411,6 +11411,9 @@ function buildClaudeArgs(opts) {
|
|
|
11411
11411
|
if (opts.settingSources) {
|
|
11412
11412
|
args.push("--setting-sources", opts.settingSources);
|
|
11413
11413
|
}
|
|
11414
|
+
if (!opts.interactive) {
|
|
11415
|
+
args.push("--permission-mode", "default");
|
|
11416
|
+
}
|
|
11414
11417
|
if (opts.jsonOutput && !opts.interactive) {
|
|
11415
11418
|
args.push("--output-format", "json");
|
|
11416
11419
|
}
|
|
@@ -29451,7 +29454,7 @@ var init_agentInstructions = __esm({
|
|
|
29451
29454
|
"src/init/agentInstructions.ts"() {
|
|
29452
29455
|
"use strict";
|
|
29453
29456
|
ARCHME_CLAUDE_BLOCK = (true ? "<!-- ARCHME-START -->\n## Architecture Awareness\n\nThis repository has a live ArchitectMe architecture connected through the `archme` MCP server.\n\nBefore broad code exploration, structural changes, new services, API/data-flow changes, or refactors:\n\n1. Use `read_architecture` or `find_components` to identify the relevant subsystem.\n2. Use `get_component` for the likely component before editing.\n3. Use `read_stories` and `check_impact` when behavior crosses component boundaries or when removing/renaming anything.\n4. Fall back to normal repo inspection if the architecture is missing, stale, or too vague.\n\nWhen working with a user on new features or large architecture changes, author a proposal (`create_proposal`) on its forked branch and let realization close it when the code lands \u2014 direct writes to the base branch are the daemon's incremental lane. See the `architecture` skill for the full flow.\n<!-- ARCHME-END -->" : fromSource("CLAUDE_BLOCK.md")).trimEnd();
|
|
29454
|
-
ARCHME_SKILL = true ? "---\nname: architecture\ndescription: Query ArchitectMe before broad code exploration, structural changes, refactors, API/data-flow work, or architecture updates. Author net-new/large design changes as proposals.\n---\n\n# Architecture Reference (ArchitectMe)\n\nThis repository's system design is managed in ArchitectMe. Use the `archme` MCP\nserver to read the live architecture while coding, and to author design changes as\nreviewable proposals. Reads and writes are scoped to the project + branch in `.mcp.json`.\n\n## Reading the architecture\n\nOrient in the map before broad repo exploration or structural edits:\n\n1. `query(\"<keyword>\")` \u2014 the one-call orientation verb. Keyword-matches components and\n returns, per match: kind/tech, a one-line description, **code anchors** (where it\n lives), direct connections, and the stories that touch it. A locator, not an\n answerer \u2014 open the anchored files yourself with Read.\n2. `read_architecture({ path: \"root\" })` \u2014 components + connections + structure at a\n scope (dot-paths: `root`, `root.api`, `root.api.db`).\n3. `get_component({ path })` \u2014 architecture + stories + context for one component.\n4. `read_stories({ path })` \u2014 user-flow stories with their Mermaid sequences.\n5. `check_impact` before removing, renaming, or moving a component boundary.\n\n**Anchors** bind a component to real code \u2014 `{ kind: \"file\" | \"directory\" | \"glob\" |\n\"symbol\", ... }`. They're how you jump from a component to its implementation: read the\nanchors, then `Read` those paths. Conceptual/external components (vendors, actors) carry\n`unanchored: true` instead. Every read surfaces a component's anchors.\n\nFor small local fixes in an already-known file, proceed normally. If ArchMe is stale,\nmissing, or too vague, fall back to regular repo inspection and mention the gap.\n\n## Proposing architecture changes (the default for net-new / large work)\n\nWhen you're about to build a net-new subsystem or make a large/structural design change,\nauthor it as a PROPOSAL \u2014 a reviewable spec against the source-of-truth branch \u2014 rather\nthan writing directly:\n\n1. `create_proposal({ title, description?, body? })` \u2014 forks a `proposal-<id>` working\n branch off the base and freezes a baseline. Returns the branch to author on, an\n `ArchMe-Proposal: <id>` trailer token, and a shareable link. **Surface the link to\n the user.** Filing a RUN of items (bugs from a triage, follow-ups from a design\n conversation)? Pass `proposals: [{ title, ... }, ...]` instead of `title` and create\n them in one call \u2014 the top-level `kind`, `parent_id`, `stage` and `base_branch` are\n defaults every item inherits and may override. Items are created in order and a\n failure part-way doesn't undo the ones before it, so retry only what the response\n reports as failed.\n Say what each one is **about** \u2014 `about: { component: 'root.\u2026' }` for the component a\n bug lives in, else `{ story: '<id>' }` for the flow, else `{ repo: 'owner/name' }`\n (null for nothing; `update_proposal` changes it later). The repo picker narrows the\n Issues and Proposals lists by it.\n2. **Author the delta on the graph \u2014 that delta IS the proposal.** On the returned\n branch, call `update_architecture` / `move_architecture` with\n `branch: \"proposal-<id>\"`: add or change the components and connections the design\n implies, and the user stories (each leaf carrying a Mermaid `sequence`) for the flows\n it introduces or changes. **Strongly prefer showing the change as a delta** \u2014 that's\n what reviewers actually read, so a representable design change described only in prose\n is close to invisible to them. Prose the graph and stories can't express \u2014 rationale,\n phased rollout, non-goals, a bug report, an ops/process change \u2014 goes in `body`. A\n prose-only proposal (populated `body`, empty delta) is valid when there's nothing\n representable to model; don't fabricate deltas to fill one out, and don't reach for\n prose to avoid modelling a change the graph could show.\n3. **Give a large proposal a walkthrough** \u2014 `set_walkthrough({ proposal: \"<id>\", steps: [...] })`.\n Each step names the stories, chapters, components and/or connections it covers,\n and the review surface filters the sequence diagram and the component graph down to\n just those, so a forty-element delta is read one bite-sized chunk at a time. Every\n step MUST reference at least one element: a step with only prose is rejected, because\n there is nothing to filter by and an explanation a reader can't check against the\n change is what this view exists to avoid. Steps go in **reading order, not build\n order** \u2014 this is not your sequencing plan restated. Titles are terse handles.\n **Every step gets a `description`**: two or three plain sentences on what this part of\n the change is and why it exists, at a high level \u2014 the narration a reviewer reads in\n guided mode while the diagrams move beside it. Not prescriptive, not a dense list of\n identifiers, and it never restates the elements the step already references.\n **Cover every change**: every story and component the delta touches must be named by\n a step, or `set_walkthrough` rejects the walkthrough and lists what is missing.\n Connections need no step of their own: they show up through the stories that draw\n them. A step that does name a connection inside a component writes it as\n `<component path>::<id>` (`daemon::session_auth->key_account`).\n Write it after the delta is in place, and skip it entirely for a small change \u2014 unless\n one already exists: when `update_architecture` warns that your writes left it out of\n date, rewrite it, or clear it with `set_walkthrough({ ..., steps: [] })`.\n4. Request review \u2014 run an automated lens yourself with `start_review` \u2192\n `submit_review` (self-contained, no user id needed), and/or `request_review` to ask a\n human teammate (needs their org-member user id).\n5. Implement in code with `ArchMe-Proposal: <id>` in the PR body
|
|
29457
|
+
ARCHME_SKILL = true ? "---\nname: architecture\ndescription: Query ArchitectMe before broad code exploration, structural changes, refactors, API/data-flow work, or architecture updates. Author net-new/large design changes as proposals.\n---\n\n# Architecture Reference (ArchitectMe)\n\nThis repository's system design is managed in ArchitectMe. Use the `archme` MCP\nserver to read the live architecture while coding, and to author design changes as\nreviewable proposals. Reads and writes are scoped to the project + branch in `.mcp.json`.\n\n## Reading the architecture\n\nOrient in the map before broad repo exploration or structural edits:\n\n1. `query(\"<keyword>\")` \u2014 the one-call orientation verb. Keyword-matches components and\n returns, per match: kind/tech, a one-line description, **code anchors** (where it\n lives), direct connections, and the stories that touch it. A locator, not an\n answerer \u2014 open the anchored files yourself with Read.\n2. `read_architecture({ path: \"root\" })` \u2014 components + connections + structure at a\n scope (dot-paths: `root`, `root.api`, `root.api.db`).\n3. `get_component({ path })` \u2014 architecture + stories + context for one component.\n4. `read_stories({ path })` \u2014 user-flow stories with their Mermaid sequences.\n5. `check_impact` before removing, renaming, or moving a component boundary.\n\n**Anchors** bind a component to real code \u2014 `{ kind: \"file\" | \"directory\" | \"glob\" |\n\"symbol\", ... }`. They're how you jump from a component to its implementation: read the\nanchors, then `Read` those paths. Conceptual/external components (vendors, actors) carry\n`unanchored: true` instead. Every read surfaces a component's anchors.\n\nFor small local fixes in an already-known file, proceed normally. If ArchMe is stale,\nmissing, or too vague, fall back to regular repo inspection and mention the gap.\n\n## Proposing architecture changes (the default for net-new / large work)\n\nWhen you're about to build a net-new subsystem or make a large/structural design change,\nauthor it as a PROPOSAL \u2014 a reviewable spec against the source-of-truth branch \u2014 rather\nthan writing directly:\n\n1. `create_proposal({ title, description?, body? })` \u2014 forks a `proposal-<id>` working\n branch off the base and freezes a baseline. Returns the branch to author on, an\n `ArchMe-Proposal: <id>` trailer token, and a shareable link. **Surface the link to\n the user.** Filing a RUN of items (bugs from a triage, follow-ups from a design\n conversation)? Pass `proposals: [{ title, ... }, ...]` instead of `title` and create\n them in one call \u2014 the top-level `kind`, `parent_id`, `stage` and `base_branch` are\n defaults every item inherits and may override. Items are created in order and a\n failure part-way doesn't undo the ones before it, so retry only what the response\n reports as failed.\n Say what each one is **about** \u2014 `about: { component: 'root.\u2026' }` for the component a\n bug lives in, else `{ story: '<id>' }` for the flow, else `{ repo: 'owner/name' }`\n (null for nothing; `update_proposal` changes it later). The repo picker narrows the\n Issues and Proposals lists by it.\n2. **Author the delta on the graph \u2014 that delta IS the proposal.** On the returned\n branch, call `update_architecture` / `move_architecture` with\n `branch: \"proposal-<id>\"`: add or change the components and connections the design\n implies, and the user stories (each leaf carrying a Mermaid `sequence`) for the flows\n it introduces or changes. **Strongly prefer showing the change as a delta** \u2014 that's\n what reviewers actually read, so a representable design change described only in prose\n is close to invisible to them. Prose the graph and stories can't express \u2014 rationale,\n phased rollout, non-goals, a bug report, an ops/process change \u2014 goes in `body`. A\n prose-only proposal (populated `body`, empty delta) is valid when there's nothing\n representable to model; don't fabricate deltas to fill one out, and don't reach for\n prose to avoid modelling a change the graph could show.\n3. **Give a large proposal a walkthrough** \u2014 `set_walkthrough({ proposal: \"<id>\", steps: [...] })`.\n Each step names the stories, chapters, components and/or connections it covers,\n and the review surface filters the sequence diagram and the component graph down to\n just those, so a forty-element delta is read one bite-sized chunk at a time. Every\n step MUST reference at least one element: a step with only prose is rejected, because\n there is nothing to filter by and an explanation a reader can't check against the\n change is what this view exists to avoid. Steps go in **reading order, not build\n order** \u2014 this is not your sequencing plan restated. Titles are terse handles.\n **Every step gets a `description`**: two or three plain sentences on what this part of\n the change is and why it exists, at a high level \u2014 the narration a reviewer reads in\n guided mode while the diagrams move beside it. Not prescriptive, not a dense list of\n identifiers, and it never restates the elements the step already references.\n **Cover every change**: every story and component the delta touches must be named by\n a step, or `set_walkthrough` rejects the walkthrough and lists what is missing.\n Connections need no step of their own: they show up through the stories that draw\n them. A step that does name a connection inside a component writes it as\n `<component path>::<id>` (`daemon::session_auth->key_account`).\n Write it after the delta is in place, and skip it entirely for a small change \u2014 unless\n one already exists: when `update_architecture` warns that your writes left it out of\n date, rewrite it, or clear it with `set_walkthrough({ ..., steps: [] })`.\n4. Request review \u2014 run an automated lens yourself with `start_review` \u2192\n `submit_review` (self-contained, no user id needed), and/or `request_review` to ask a\n human teammate (needs their org-member user id).\n5. Implement in code with one `ArchMe-Proposal: <id>` line in the PR body per proposal or\n issue the PR implements or fixes. Merging closes each, plus the issues those proposals\n resolve. **Never merge the proposal branch into the base branch yourself.**\n\nUse `add_proposal_comment` to discuss a proposal **or a pull request** \u2014 free-form,\noptionally anchored to a component/story/connection (`anchor_kind` + `anchor_id`) or left\nas an entity-level thread, and `in_reply_to` to thread replies. Pass `proposal_id` OR\n`pr`, never both; a `pr` also takes `repo` (`\"owner/name\"`) when two watched repos in the\nproject share its number \u2014 the error names them. Distinct from `submit_review` (a typed lens verdict): use comments for\nconversation and questions on the delta.\n\nRead a PR back with `get_pr` \u2014 GitHub's own facts about it, ArchMe's description, its\nwalkthrough, and the whole comment thread. That is how you see a REPLY to a comment you\nleft, so check it before assuming a question went unanswered.\n\nA comment on a PR (or on a proposal some PR's `ArchMe-Proposal:` trailer names) is\nmirrored to that pull request on GitHub, so reviewers who live there see it. Write for\nthat audience: the first sentence is what they read.\n\n## Updating the architecture (update_architecture)\n\n`update_architecture` is additive \u2014 it merges components, connections, stories, and\nrequirements by id. Use it to author a proposal's spec on the fork; direct writes to the\nbase branch are reserved for the daemon's small incremental diffs.\n\nKeep the graph honest:\n\n- **Every component is anchored** to the code that implements it (`anchors`), or\n explicitly `unanchored: true` if it's conceptual/external. An unanchored internal\n component is drift.\n- **Connections reflect real data flow** \u2014 every call/dependency between components is a\n connection (`from`, `to`, `method`). Name the endpoints with `root.` paths; ArchMe\n stores each connection where its endpoints meet and the result says where (scope + id).\n- Add children with a separate entry using `at` (don't nest a `components` array).\n- Stories capture user flows; each leaf story carries a Mermaid `sequence` whose\n participants are `root.`-prefixed dot-paths.\n\nUse `move_architecture` to relocate components/stories (it rewrites sequences +\nconnections) and `update_node_context` for design decisions, references, and notes.\n\n## Scan\n\nUse the `/scan` MCP prompt to bootstrap or refresh the architecture from the local repo.\nFollow its setup, branch, and restart instructions.\n\n## Common Mistakes\n\n- Do not grep/read broadly first when a `query` or architecture lookup can narrow it.\n- Do not push net-new or large changes directly to the base branch \u2014 open a proposal.\n- Do not settle for a prose-only proposal when it adds or changes components,\n connections, or user flows \u2014 model those on the fork branch; `body` is for what the\n graph can't show.\n- Do not leave internal components unanchored or real connections missing.\n- Do not assume the architecture is complete; verify against code (via anchors) before\n risky edits.\n- Do not remove or rename architectural boundaries without `check_impact`.\n- Do not merge a proposal branch into the base \u2014 realization closes it when code lands.\n" : fromSource("architecture-skill.md");
|
|
29455
29458
|
}
|
|
29456
29459
|
});
|
|
29457
29460
|
|
package/package.json
CHANGED
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
{
|
|
2
2
|
"name": "@archme/cli",
|
|
3
|
-
"version": "0.25.
|
|
3
|
+
"version": "0.25.1",
|
|
4
4
|
"description": "Map your codebase into a living architecture diagram: run `archme init` in a repo to scan it, `archme daemon start` to keep it current on every commit.",
|
|
5
5
|
"type": "module",
|
|
6
6
|
"license": "SEE LICENSE IN LICENSE",
|
package/sandbox-image.json
CHANGED