Tags: chriswritescode-dev/opencode-forge
Tags
fix: stop guarding question tool on captured plan The tool.execute.before hook used to capture the latest assistant plan and throw 'Forge plan capture failed' when no plan existed, blocking any question tool call whose options included an 'Execute here' label. That detection was too loose (single-label match) and the guard itself was unnecessary: the post-answer routing in tool.execute.after already no-ops when the answer does not match an execution label, and refuses to dispatch when no plan is found. - Drop the pre-execution plan capture and throw path entirely. - Tighten isPlanApprovalQuestionArgs to require all three canonical labels (New session, Execute here, Loop) before treating a question as plan approval. - Remove now-unused imports and helpers.
refactor: separate bash vs forge-bash permissions for sandbox loops (#30 ) * loop: bash-tool-parity removed after 0 iterations * refactor: separate bash vs forge-bash permissions for sandbox loops * refactor: remove tree-sitter-bash dependency and consolidate bash tool parsing * refactor: add sandbox option to loop and audit permission rulesets * chore: bump version to 0.4.7
fix: clarify forge-section marker placement in architect prompts (#26) * loop: no-boot-recovery completed after 17 iterations * fix: clarify forge-section marker placement in architect prompts * fix: add missing status field to LoopState objects in tests * chore: bump version to 0.4.5
Snapshot of release/v0.3.0 prior to further iteration Branch: release/v0.3.0 Commit: c9babf8 (refactor: consolidate loop hooks implementation and tests) Purpose: known-good baseline; iterate freely and return here if needed. Restore with: git checkout v0.3.0-rc1
PreviousNext