fix(workflows,commands): preserve repo identity for issue URLs; unblock .archon fixes (#2417)

Author: Wirasm

.archon/commands/defaults/archon-fix-issue.md

@@ -20,10 +20,47 @@ created a git worktree on the correct branch. In that case:
   operator's `.archon/` directory — workflows, commands, scripts — into every run
   worktree, deliberately, so a workflow can be iterated on before it is committed.
   Those files are present *before* you start and are not your changes.
-- **Modifications under `.archon/` are never yours to commit, stash, or remove.**
-  Leave them exactly as they are and commit only the files your implementation touched.
-  Before every commit, confirm with `git diff --cached --name-only` that nothing under
-  `.archon/` is staged.
+- **Pre-existing modifications under `.archon/` are never yours to commit, stash, or
+  remove.** Leave them exactly as they are and commit only the files your implementation
+  touched. Before every commit, confirm with `git diff --cached --name-only` that no
+  `.archon/` file you did not deliberately change is staged.
+- **The exception: when the issue's fix genuinely lives under `.archon/`.** Workflows,
+  commands and scripts are source too, and an issue can legitimately target one. If your
+  plan says to edit a specific `.archon/` file, edit and commit **that file** — the rule
+  above exists to stop you sweeping up the operator's unrelated copied-in edits, not to
+  make a whole directory unfixable.
+
+  Distinguish the two by intent, not by path: a file your plan names is your work; every
+  other dirty `.archon/` file is not. On 2026-08-03 a run blocked outright on this,
+  correctly reporting "contradictory instructions" because the issue required editing a
+  workflow YAML while this section forbade touching anything under `.archon/`. It was
+  right to refuse rather than guess — and the rule was wrong to be absolute.
+
+  **A named file is not a blank cheque for that file.** It may already carry copied-in
+  edits from before you started, and staging it whole would commit those too — the
+  path-level check above cannot see inside a file. So before you touch a planned
+  `.archon/` file, record its baseline:
+
+  ```bash
+  # HEAD, not the index: `git diff -- <file>` compares the worktree against the
+  # INDEX, so pre-existing changes that are already STAGED do not appear — and
+  # `git add -p` will neither show nor remove them, so they ride into your commit
+  # invisibly. Diffing against HEAD captures staged and unstaged alike.
+  git diff HEAD -- <the-planned-file> > /tmp/archon-baseline.diff   # empty if clean
+  ```
+
+  Then, before staging anything of your own, clear that file out of the index so the
+  only thing you can stage is what you deliberately pick:
+
+  ```bash
+  git restore --staged <the-planned-file>   # no-op if nothing was staged
+  git add -p <the-planned-file>             # stage ONLY your own hunks
+  ```
+
+  Reject any hunk that also appears in the baseline. If yours and theirs are entangled
+  such that you cannot separate them, stop and say so rather than committing someone
+  else's work under your change. That is the same call the 2026-08-03 run made, and it
+  was the right one.
 - **Dirty paths outside `.archon/` are also not a reason to stop, and also not yours.**
   They are either your own work from an earlier attempt at this run (resume reuses the
   worktree) or something the operator left behind. Either way: leave them alone, do not

.archon/workflows/defaults/archon-fix-github-issue.yaml

@@ -31,22 +31,55 @@ nodes:
 
       Request: $ARGUMENTS
 
-      Rules:
-      - If the message contains an explicit issue number (e.g., "#709", "issue 709", "709"), extract that number.
-      - If the message is ambiguous (e.g., "fix the SQLite timestamp bug"), use `gh issue list` to search for matching issues and pick the best match.
-
-      CRITICAL: Your final output must be ONLY the bare number with no quotes, no markdown, no explanation. Example correct output: 709
+      A reference is an ATOMIC pair: a number, and the repo it belongs to. Never take the
+      number from one reference and the repo from another — that combination fetches an
+      issue nobody asked for.
+
+      Recognised reference forms, each yielding one complete pair:
+      - `#709`, `issue 709`, `709`            -> {issue_number: "709", repo: ""}
+      - `owner/repo#709`                      -> {issue_number: "709", repo: "owner/repo"}
+      - `https://github.com/owner/repo/issues/709`
+                                              -> {issue_number: "709", repo: "owner/repo"}
+
+      An empty `repo` means "the current checkout" — the common case. NEVER guess a repo.
+
+      Then:
+      - Exactly one reference -> use it.
+      - Several references, all resolving to the SAME pair -> use it.
+      - Several DIFFERENT references (e.g. a bare number and a cross-repo URL) -> the
+        request is ambiguous. Do NOT pick one. Fail with a message naming every pair you
+        found and asking which was meant. Guessing here silently investigates the wrong
+        issue, which is worse than stopping.
+      - No reference at all (e.g. "fix the SQLite timestamp bug") -> search the CURRENT
+        repo with `gh issue list`, pick the best match, and leave `repo` empty.
+    output_format:
+      type: object
+      properties:
+        issue_number:
+          type: string
+        repo:
+          type: string
+      required:
+        - issue_number
+        - repo
 
   - id: fetch-issue
     bash: |
-      # $extract-issue-number.output is injected pre-quoted by Archon — do NOT
-      # wrap it in double quotes (that corrupts the value; see issue #1884).
-      ISSUE_NUM=$(echo $extract-issue-number.output | grep -oE '[0-9]+' | head -1)
-      if [ -z "$ISSUE_NUM" ]; then
-        echo "Failed to extract issue number from:" $extract-issue-number.output >&2
+      # Substitutions are injected already shell-quoted by Archon — assign them
+      # unquoted, then quote normally as locals (see #1884).
+      num=$extract-issue-number.output.issue_number
+      repo=$extract-issue-number.output.repo
+      if ! printf '%s' "$num" | grep -qE '^[0-9]+$'; then
+        echo "extract-issue-number did not yield a numeric issue id: $num" >&2
         exit 1
       fi
-      gh issue view "$ISSUE_NUM" --json title,body,labels,comments,state,url,author
+      # An explicit owner/repo must be preserved: `gh issue view <n>` resolves against
+      # the CURRENT checkout, so a cross-repo URL would silently read this repo's #<n>.
+      if [ -n "$repo" ]; then
+        gh issue view "$num" --repo "$repo" --json title,body,labels,comments,state,url,author
+      else
+        gh issue view "$num" --json title,body,labels,comments,state,url,author
+      fi
     depends_on: [extract-issue-number]
 
   - id: classify

.archon/workflows/experimental/archon-fix-github-issue-experimental.yaml

@@ -125,22 +125,56 @@ nodes:
 
       Request: $ARGUMENTS
 
-      Rules:
-      - If the message contains an explicit issue number (e.g., "#709", "issue 709", "709"), extract that number.
-      - If the message is ambiguous (e.g., "fix the SQLite timestamp bug"), use `gh issue list` to search for matching issues and pick the best match.
-
-      CRITICAL: Your final output must be ONLY the bare number with no quotes, no markdown, no explanation. Example correct output: 709
+      A reference is an ATOMIC pair: a number, and the repo it belongs to. Never take the
+      number from one reference and the repo from another — that combination fetches an
+      issue nobody asked for.
+
+      Recognised reference forms, each yielding one complete pair:
+      - `#709`, `issue 709`, `709`            -> {issue_number: "709", repo: ""}
+      - `owner/repo#709`                      -> {issue_number: "709", repo: "owner/repo"}
+      - `https://github.com/owner/repo/issues/709`
+                                              -> {issue_number: "709", repo: "owner/repo"}
+
+      An empty `repo` means "the current checkout" — the common case. NEVER guess a repo.
+
+      Then:
+      - Exactly one reference -> use it.
+      - Several references, all resolving to the SAME pair -> use it.
+      - Several DIFFERENT references (e.g. a bare number and a cross-repo URL) -> the
+        request is ambiguous. Do NOT pick one. Fail with a message naming every pair you
+        found and asking which was meant. Guessing here silently investigates the wrong
+        issue, which is worse than stopping.
+      - No reference at all (e.g. "fix the SQLite timestamp bug") -> search the CURRENT
+        repo with `gh issue list`, pick the best match, and leave `repo` empty.
+    output_format:
+      type: object
+      properties:
+        issue_number:
+          type: string
+        repo:
+          type: string
+      required:
+        - issue_number
+        - repo
     model: small
 
   - id: fetch-issue
     bash: |
-      # Strip quotes, whitespace, markdown backticks from AI output
-      ISSUE_NUM=$(echo "$extract-issue-number.output" | tr -d "'\"\`\n " | grep -oE '[0-9]+' | head -1)
-      if [ -z "$ISSUE_NUM" ]; then
-        echo "Failed to extract issue number from: $extract-issue-number.output" >&2
+      # Substitutions are injected already shell-quoted by Archon — assign them
+      # unquoted, then quote normally as locals (see #1884).
+      num=$extract-issue-number.output.issue_number
+      repo=$extract-issue-number.output.repo
+      if ! printf '%s' "$num" | grep -qE '^[0-9]+$'; then
+        echo "extract-issue-number did not yield a numeric issue id: $num" >&2
         exit 1
       fi
-      gh issue view "$ISSUE_NUM" --json title,body,labels,comments,state,url,author
+      # An explicit owner/repo must be preserved: `gh issue view <n>` resolves against
+      # the CURRENT checkout, so a cross-repo URL would silently read this repo's #<n>.
+      if [ -n "$repo" ]; then
+        gh issue view "$num" --repo "$repo" --json title,body,labels,comments,state,url,author
+      else
+        gh issue view "$num" --json title,body,labels,comments,state,url,author
+      fi
     depends_on: [extract-issue-number]
 
   - id: classify