[Question/Bug] archive without syncing #764

Activity

1. 21st-connector commented on Feb 26, 2026

   

   21st-connectorBot

   on Feb 26, 2026 – with 21st - Connector

   Last edited by 21st-connector

   Contributor

   More actions

   Task completed.

   The enrichment comment has been posted on issue #764. Here's a summary of what was included:

   Key findings:

   • The sync-before-archive prompt is entirely agent-driven via skill template instructions (src/core/templates/workflows/archive-change.ts:63-67), not deterministic code — which is why different agents/models handle it inconsistently.
   • The sync workflow is not in the core profile (src/core/profiles.ts:14), so core profile users who skip sync at archive time have no recovery path.
   • The separation was intentional (PRs feat: add /opsx:sync command for syncing delta specs to main specs #450, feat: add smart sync check to /opsx:archive command #452, Refine opsx archive sync assessment #514, fix: use Skill tool for sync invocation in archive templates #632), but a previous bug (fix: use Skill tool for sync invocation in archive templates #632) showed agents were already struggling to invoke sync correctly during archive.
   • No duplicate issues exist, though question: why achive skill or command relies on llm to manually merge specs why it does not use openspec archive cli command? #656 and Best workflow to fix mistakes in a completed change? #618 are related.

   Three potential approaches were suggested:

   1. Always sync before archive (as proposed), with an opt-out --skip-sync flag
   2. Move sync-at-archive into deterministic CLI code rather than agent instructions
   3. Add sync to the core profile as a minimum recovery path

   ---

   View full conversation

   Powered by 1Code
2. harikrishnan83

   changed the title [-]Archive without syncing[/-] [+][Question/Bug] archive without syncing[/+] on Feb 26, 2026
3. 21st-connector commented on Feb 26, 2026

   

   21st-connectorBot

   on Feb 26, 2026 – with 21st - Connector

   Contributor

   More actions

   Issue Analysis

   Thanks for the detailed report. Here's a codebase analysis to help contextualize this request.

   Relevant Code

   The sync-before-archive prompt is defined in the archive skill template:

   • src/core/templates/workflows/archive-change.ts:63-67 — The prompt logic:

     • If delta spec changes are needed: offers "Sync now (recommended)" vs "Archive without syncing"
     • If already synced: offers "Archive now", "Sync anyway", "Cancel"

   • src/core/templates/workflows/archive-change.ts:56-67 — The full sync assessment step: checks for delta specs at openspec/changes/<name>/specs/, compares each with its corresponding main spec, then shows a combined summary before prompting.

   • src/core/templates/workflows/archive-change.ts:67 — If sync is chosen, it invokes openspec-sync-specs via the Skill tool (a sub-agent). The key line: "Proceed to archive regardless of choice" — this means sync is always optional from the archive flow's perspective.

   The same prompt structure is duplicated in the OPSX command template at line 179-180 of the same file.

   Core vs. Custom Profile

   The issue author correctly identifies that sync is not in the core profile:

   • src/core/profiles.ts:14 — Core profile: ['propose', 'explore', 'apply', 'archive']
   • src/core/profiles.ts:19-30 — sync is in ALL_WORKFLOWS but only available when using the custom profile

   This means core profile users who skip sync at archive time have no standalone /opsx:sync command to run afterward — the delta specs are archived away and can no longer be applied to main specs without manual extraction.

   The Core Implementation

   The CLI's ArchiveCommand class (src/core/archive.ts:50) handles the actual move operation but does not contain sync logic itself. Spec application is in src/core/specs-apply.ts (used by the CLI's --skip-specs flag path). The sync-at-archive-time decision is entirely agent-driven via the skill template instructions.

   Design History

   The separation of sync from archive was an intentional design choice introduced in these PRs:

   • PR feat: add /opsx:sync command for syncing delta specs to main specs #450 — Added /opsx:sync as a standalone command, decoupling sync from archive so specs can be synced anytime during implementation
   • PR feat: add smart sync check to /opsx:archive command #452 — Added "smart sync check" to archive — examines delta specs vs main specs and only prompts when sync is actually needed
   • PR Refine opsx archive sync assessment #514 — Refined the sync assessment to show per-delta-spec evaluation with a combined summary
   • PR fix: use Skill tool for sync invocation in archive templates #632 — Fixed a bug where agents couldn't properly invoke sync during archive (changed from slash command to Skill tool invocation). This is directly relevant — before this fix, agents would improvise sync behavior incorrectly

   The Problem Pattern

   The issue describes a real gap in the current design:

   1. Agent reliability: The sync prompt is handled by the AI agent interpreting skill template instructions, not by deterministic code. Some models/agents may skip the prompt or default to "archive without syncing" — especially in non-interactive or fast-path scenarios.
   2. No recovery path on core profile: If sync is skipped at archive time and the user is on the core profile, delta specs are moved into the archive directory with no standalone sync command available to recover them.
   3. Silent data loss: The user may not realize sync was skipped, particularly if the agent doesn't clearly surface the choice.

   Potential Approaches

   Based on the codebase, here are a few paths forward:

   1. Make sync-before-archive the default (as proposed): Remove the "Archive without syncing" option and always sync delta specs before archiving. This would mean modifying the skill templates in src/core/templates/workflows/archive-change.ts (both the skill and command templates) and potentially adding a --skip-sync CLI flag for power users who explicitly want to opt out.

   2. Make sync deterministic in the CLI: Move the sync-at-archive logic from the agent-driven skill template into the ArchiveCommand class in src/core/archive.ts, so it runs as code rather than relying on agent interpretation. The CLI already has specs-apply.ts for this purpose — the --skip-specs flag path uses it.

   3. Add sync to the core profile: Change CORE_WORKFLOWS in src/core/profiles.ts:14 to include 'sync', so core profile users always have a recovery path via /opsx:sync even if they skip at archive time.

   Related Issues

   • question: why achive skill or command relies on llm to manually merge specs why it does not use openspec archive cli command? #656 — Questions why archive skill relies on LLM to manually merge specs (related concern about agent-driven spec operations)
   • Best workflow to fix mistakes in a completed change? #618 — Asks about workflow for fixing mistakes in a completed change (relevant to the recovery gap)
   • No existing issue duplicates this request.
4. harikrishnan83 commented on May 1, 2026

   

   harikrishnan83

   on May 1, 2026

   ContributorAuthor

   More actions

   With sync being part of the core workflow in #1030 this issue resolved. Thanks @TabishB .
5. harikrishnan83

   closed this as completedon May 1, 2026