Repository navigation
Conversation
…ctice Packs (#520) Records what Archon v0.12.1 offers a third party that wants to add or replace steps in a workflow it does not own: prompt-only templating, include slots shadowed by filename across scopes, no project-level skills default, and the 0.12.0 per-node plugin scoping. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Archon 0.12.0 stopped Claude nodes loading installed Claude Code plugins unless the node names them in plugins:. The 2026-10-01 line said nodes load user-scope plugins by default, which now holds only up to 0.11. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…workflow-extensibility docs(unic-archon-dlc): research Archon workflow extensibility for Practice Packs
Several Practice Packs may share one Consumer with the DLC. ADR-0038 records the eight decisions from the grilling on #520: ownership by prefix, the DLC alone owns .archon/methods/, layered config, and the include: slot as the unbuilt route to replace a Box step. Each Archon fact names the version it was read or measured on. ADR-0031, ADR-0024 and ADR-0013 point at it. CONTEXT.md states the Practice Pack model as decided. Refs #520 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Under ADR-0038 every installer owns only the names that carry its prefix, and the DLC's prefix is unic-dlc. /setup now writes its blocks between unic-dlc:begin and unic-dlc:end. It finds a block under the old unic-archon-dlc markers and rewrites it in place, so a re-run leaves one block. A begin marker without an end marker stops the write to that file. /setup also states that it touches no name carrying another prefix, that it alone owns .archon/methods/, and where the route to replace a Box step is recorded. Refs #520 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…e diagram The architecture diagram labelled the Practice Pack "design open". ADR-0038 decides it, so the node and the HOW card now say a pack works under its own prefix. HTML and both PNGs are exported again. Refs #520 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The build, pr-review and qa Boxes computed expected_base from project.branching, so no run could target a branch other than develop or main. The build Box also read $BASE_BRANCH in its verify node, so one Box held two answers to the same question. Every base-reading line now reads $BASE_BRANCH. The pr-review diff reads the PR's own target branch when there is a PR, because that is the diff that will merge. project.branching keeps one job: whether /qa deletes the merged feature branch. Refs #505 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…in /setup ADR-0038 records why a run's base branch is Archon's $BASE_BRANCH and not a DLC key, what project.branching still does, and why the review diff reads the PR's target. ADR-0023's code-review row now names $BASE_BRANCH, revised inline, with its old amendment blocks folded into the body. /setup reads worktree.baseBranch and the remote default, reports the base Archon will use next to the remote line, and warns with the line to add when the key is unset or disagrees with branching. It still writes nothing to .archon/config.yaml. The README and the three Box docs no longer state the base rule. Refs #505 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The prefix rule told /setup to leave every other marked block alone, which forbade the migration Step 7 performs. It now reserves the legacy markers to this Plugin, names the full paths, says /setup writes nothing in .archon/packs/ until #647 lands, and names the unic-dlc-e2e overlap as an open risk instead of claiming the rule holds for any boundary. Step 7 checks malformed markers first: a begin without a matching end, an end without a begin, or two blocks under one name. It then writes nothing to that file and reports ACTION REQUIRED. An old block in an ignore file no formatter uses is reported too. Refs #520 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The layered config and the defaults file are #647's to build, so ADR-0038 and CONTEXT.md now say so. The DLC's prefix rule covers the four places it names, not every file /setup writes. A Consumer choice on an include slot is a choice, not a default. ADR-0013 says that the ROADMAP.md block it quotes is not migrated. Colons and passive voice are split into plain sentences. Refs #520 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The maintainer chose the docs/adr/README.md rule over amendment blocks for #520. ADR-0031 and ADR-0024 now read as current, with what still holds from their old blocks carried into Context and Decision and the blocks deleted. Their Status lines and index rows use the revised form. Citations elsewhere in the plugin that named one of those blocks or its date now cite the ADR. The Box workflow files keep their "(ADR-0031, amended)" text: it is still true, and Box edits wait for #502. Refs #520 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Step 7 now treats more than one begin or end under one name as malformed, says /setup writes nothing in .claude/skills/ today, and reports any file outside the patched set that holds an old begin or end marker without guessing why. The "Both" case names its Step 8 values. ADR-0038 says an include target resolves by name: and is overridden by filename, scopes the "only files a pack writes" sentences to v1, names the include slot as a future exception to the §5 table, names the two docs/agents files, and states the limit of the pointer measurement that #648 depends on. Refs #520 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…osed on base checks prep in /pr-review could pick a diff base that this clone never fetched. Both merge-base forms then failed, and the HEAD~1 fallback reviewed one commit and reported it clean. prep now fetches the diff base, a new guard-no-diff-base node cancels the run when it still does not resolve, and the PR-target warning is a prep output that the summary shows. verify-pr-base in /qa set base_ok to true when the tracker query failed or the tracker contract was missing. It now sets false in those cases, so merge refuses. The stub scan in /build tries origin/$BASE_BRANCH first. The approval messages name the branch through node outputs, and the bootstrap text no longer claims the worktree is always cut from the base. Refs #505 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
/setup read the remote's default branch as if Archon always stored it. That holds only from Archon 0.12.0. Archon 0.7 to 0.11 store the branch checked out at the first registration. The report now labels the value as an approximate stand-in, the unset-key warning is the main signal, an empty value prints "none resolved", and the config block prints once. ADR-0038 matches, and both ADRs lose a colon connector, a passive clause and an aside. Refs #505 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…line diagram The new cancel node made the diagram stale. It is drawn the way guard-not-ready is, as a cancel node with a security edge from prep, and gets its own "No diff base" view. The HTML is regenerated with archify deliver, and both PNGs are exported from the viewer and scaled to 2620 px wide, as before. Refs #505 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
review and guard-no-diff-base both depended only on prep, so review could start while the guard cancelled the run. review now runs only when base_ref is set, which makes its own empty-base stop unreachable, so that text is gone. The /qa merge-gate message reads correctly when no PR exists: it names the PR target, which is empty then, and points at the bootstrap message for the run base. Refs #505 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
… ADR-0038 §6 said a later layer wins, but nothing orders two packs' defaults files. The maintainer has not decided that order, so §6, the collision table and "Not decided here" now say it is open and that #647 decides it before a second pack sets a default. Refs #520 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…before every merge-base prep treated a failed tracker query like "no PR exists" and fell back to $BASE_BRANCH. It now blocks with an empty base_ref, so the guard cancels. Only a query that confirms no PR falls back. prep prints a failed fetch and the ref that resolved, and a blocked prep still writes prep.json. The guard's cancel message says what to fix in each case. /qa's merge-gate embeds the bootstrap message, and verify-pr-base prints its own line for the no-PR and failed-query cases. /build's code-review pre-check fetches the base before it takes the merge-base. Refs #505 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…as no PR A tracker contract may name no server that lists pull requests. That is a stated fact, not a failed query, so prep continues as for a confirmed no-PR case and diffs against $BASE_BRANCH. Refs #505 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
ADR-0024's title now names the tracker contract in docs/agents/, and the false note about its filename is gone. ADR-0019 says the label rules were recorded in an ADR-0024 amendment that the inline revision removed. ADR-0038 says the in-scope duplicate-name result follows from the cross-scope measurement. ADR-0032 drops "amended", and three mid-sentence colons become sentences. Refs #520 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…e-packs-stack feat(unic-archon-dlc): Practice Packs stack by prefix (ADR-0038), markers move to unic-dlc
… base claims ADR-0038 now cites the re-run of spike sub-item 4 on the current Box, and says that the cancel path of guard-no-diff-base is unmeasured and that #502 proves it in the Consumer. /setup prints "Verified" only for worktree.baseBranch, quotes the v0.12.0 release note for how new projects pick their default branch, and no longer claims to be the only place that names a main line, since /cleanup keeps its hint. AGENTS.md and the /pr-review diagram name guard-no-diff-base. The diagram's HTML is regenerated. Its PNGs come out byte-identical, because the changed view note is not drawn on the default canvas. Refs #505 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…n explicit refspec In a single-branch clone, `git fetch origin <branch>` writes only FETCH_HEAD, so origin/<branch> stays unresolved and prep's guard cancels after a fetch that succeeded. prep and /build's code-review pre-check now fetch +refs/heads/<branch>:refs/remotes/origin/<branch>, which works on git 2.25. Refs #505 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…anch-from-archon feat(unic-archon-dlc): read the run's base branch from $BASE_BRANCH
…ts file changed The next review omitting a finding is not evidence that it is gone. On a byte-identical commit, run 3 ruled six findings fixed and resolved their threads. Each iteration now records the commit it reviewed in a hidden head marker in its summary, written from `git rev-parse HEAD`. reconcile writes `git diff --name-only` between the prior head and this head to the session directory, and rules an omitted prior finding `fixed` only when its file is in that list. Otherwise the finding is `unverified`. post replies once on an `unverified` thread and leaves its status. It reads each thread's status first, reopens a resolved thread whose finding is back, and skips a reply when the thread's latest reply marker already holds the same hash and verdict. `unverified_fixed` is replaced by `unverified`. Closes #482 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This branch was previously deployed
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.
Why
mainhas been at the 0.29.0 state ofunic-archon-dlcsince 2026-10-19. That plugin is now deprecated and out of the catalogue (#677, #679). This release bringsmainlevel withdevelop, so thatmainno longer differs fromdevelopin work nobody will release.What
From
develop, through these PRs:unic-archon-dlc, 0.29.0 to 0.36.1 on disk: docs(unic-archon-dlc): research Archon workflow extensibility for Practice Packs #646, feat(unic-archon-dlc): Practice Packs stack by prefix (ADR-0038), markers move to unic-dlc #649, feat(unic-archon-dlc): read the run's base branch from $BASE_BRANCH #650, fix(unic-archon-dlc): rule an omitted prior finding fixed only when its file changed #661, fix(unic-archon-dlc): end every /qa run that does not merge cancelled #663, feat(unic-archon-dlc)!: open no pull request on a verdict other than pass #665, feat(unic-archon-dlc): gate the build verdict on the pre-check #668, fix(unic-archon-dlc): attribute reported checks in the pr-review summary #669 and fix(unic-archon-dlc): remove references back to this repository from what a Consumer reads #673. The plugin stays out of.claude-plugin/marketplace.json, so neither catalogue offers it.develop's side.unic-archon-dlcfrom the release tag list (Stop tagging unic-archon-dlc releases on main #684).mainin the README install command..claude/rules/learned.mdand enableunic-learning-loopin the committed settings.No other plugin changes.
mainanddevelopcarry the same version of every other plugin.After the merge
unic-archon-dlcis no longer on the tag list.unic-archon-dlc@0.29.0stays its last tag.unchanged 7andremoved 0.🤖 Generated with Claude Code