[AUTO-EVLV] Proposal: Fix `forgejo_issue_add_comment` → `forgejo_create_issue_comment` permission name mismatch (17 agents) #8404

New issue

Open

opened 2026-04-13 18:40:28 +00:00 by HAL9000 · 1 comment

HAL9000 commented

2026-04-13 18:40:28 +00:00

Owner

Agent Evolution Proposal — Cycle 1

Category: Task-type failures (agents silently failing to post comments)
Severity: High — breaks escalation tracking system
Affected files: 17 agent definition files in .opencode/agents/

Problem

Across 17 agent definition files, the Forgejo permission entry reads:

"forgejo_issue_add_comment": allow

However, the actual MCP tool name for posting issue/PR comments is:

"forgejo_create_issue_comment": allow

Because the permission system uses exact tool name matching, any agent that tries to call forgejo_create_issue_comment will be silently denied — the tool call will fail with a permission error, even though the agent's YAML frontmatter appears to grant comment access.

Evidence

The correct tool is confirmed by the MCP tool definition:

forgejo_create_issue_comment — Create issue comment

The wrong name forgejo_issue_add_comment appears in 17 agent files (confirmed via grep):

Agent File	Impact
`implementation-worker.md`	Cannot leave attempt comments → breaks escalation tracking
`implementation-pool-supervisor.md`	Cannot post escalation notices
`pr-merge-pool-supervisor.md`	Cannot post merge verification comments
`human-liaison-worker.md`	Cannot communicate with humans via comments
`human-liaison-pool-supervisor.md`	Cannot post liaison status comments
`epic-planning-worker.md`	Cannot update epic planning comments
`epic-planning-pool-supervisor.md`	Cannot post planning status
`grooming-worker.md`	Cannot post grooming notes
`system-watchdog-worker.md`	Cannot explain corrections made
`new-issue-creator.md`	Cannot post creation confirmations
`issue-note-writer.md`	Cannot write issue notes
`milestone-reviewer.md`	Cannot post review comments
`project-owner-worker.md`	Cannot post ownership updates
`project-owner-pool-supervisor.md`	Cannot post supervisor status
`spec-update-pool-supervisor.md`	Cannot post spec update status
`state-reconciler.md`	Cannot post reconciliation notes
`subtask-loop.md`	Cannot post subtask progress

Most critical impact: The implementation-worker.md agent is explicitly designed to leave attempt comments after every task:

"These comments are how you track escalation state across worker sessions."

Without the ability to post comments, the entire progressive escalation system (Tier 1→4) is broken — the supervisor cannot read previous attempt comments to determine the correct escalation tier.

Proposed Change

In all 17 affected files, replace:

"forgejo_issue_add_comment": allow

with:

"forgejo_create_issue_comment": allow

This is a pure permission name correction — no logic changes, no behavioral changes. The agents already intend to post comments; this fix makes the permission actually grant that capability.

Verification

After the fix, implementation workers should be able to leave attempt comments on issues and PRs, enabling the supervisor to correctly track escalation state across worker sessions.

Please approve this proposal by:

Removing the needs feedback label, OR
Adding the State/Verified label, OR
Leaving a comment with explicit approval

Automated by CleverAgents Bot
Supervisor: Agent Evolution | Agent: agent-evolution-pool-supervisor

## Agent Evolution Proposal — Cycle 1 **Category**: Task-type failures (agents silently failing to post comments) **Severity**: High — breaks escalation tracking system **Affected files**: 17 agent definition files in `.opencode/agents/` --- ## Problem Across 17 agent definition files, the Forgejo permission entry reads: ```yaml "forgejo_issue_add_comment": allow ``` However, the **actual MCP tool name** for posting issue/PR comments is: ```yaml "forgejo_create_issue_comment": allow ``` Because the permission system uses exact tool name matching, any agent that tries to call `forgejo_create_issue_comment` will be **silently denied** — the tool call will fail with a permission error, even though the agent's YAML frontmatter appears to grant comment access. --- ## Evidence The correct tool is confirmed by the MCP tool definition: > `forgejo_create_issue_comment` — Create issue comment The wrong name `forgejo_issue_add_comment` appears in **17 agent files** (confirmed via grep): | Agent File | Impact | |---|---| | `implementation-worker.md` | Cannot leave attempt comments → **breaks escalation tracking** | | `implementation-pool-supervisor.md` | Cannot post escalation notices | | `pr-merge-pool-supervisor.md` | Cannot post merge verification comments | | `human-liaison-worker.md` | Cannot communicate with humans via comments | | `human-liaison-pool-supervisor.md` | Cannot post liaison status comments | | `epic-planning-worker.md` | Cannot update epic planning comments | | `epic-planning-pool-supervisor.md` | Cannot post planning status | | `grooming-worker.md` | Cannot post grooming notes | | `system-watchdog-worker.md` | Cannot explain corrections made | | `new-issue-creator.md` | Cannot post creation confirmations | | `issue-note-writer.md` | Cannot write issue notes | | `milestone-reviewer.md` | Cannot post review comments | | `project-owner-worker.md` | Cannot post ownership updates | | `project-owner-pool-supervisor.md` | Cannot post supervisor status | | `spec-update-pool-supervisor.md` | Cannot post spec update status | | `state-reconciler.md` | Cannot post reconciliation notes | | `subtask-loop.md` | Cannot post subtask progress | **Most critical impact**: The `implementation-worker.md` agent is explicitly designed to leave attempt comments after every task: > "These comments are how you track escalation state across worker sessions." Without the ability to post comments, the entire progressive escalation system (Tier 1→4) is broken — the supervisor cannot read previous attempt comments to determine the correct escalation tier. --- ## Proposed Change In all 17 affected files, replace: ```yaml "forgejo_issue_add_comment": allow ``` with: ```yaml "forgejo_create_issue_comment": allow ``` This is a pure permission name correction — no logic changes, no behavioral changes. The agents already intend to post comments; this fix makes the permission actually grant that capability. --- ## Verification After the fix, implementation workers should be able to leave attempt comments on issues and PRs, enabling the supervisor to correctly track escalation state across worker sessions. --- **Please approve this proposal by:** 1. Removing the `needs feedback` label, OR 2. Adding the `State/Verified` label, OR 3. Leaving a comment with explicit approval --- **Automated by CleverAgents Bot** Supervisor: Agent Evolution | Agent: agent-evolution-pool-supervisor

HAL9000 commented

2026-04-13 18:48:14 +00:00

Author

Owner

⚠️ Awaiting Human Approval — This proposal requires human review before implementation.

This issue was created by the Agent Evolution Supervisor (AUTO-EVLV) as a Step 1 Proposal. No code changes have been made yet.

To approve this proposal, please do one of the following:

Remove the needs feedback label from this issue
Add the State/Verified label
Leave a comment with explicit approval (e.g., "approved" or "LGTM")

To reject this proposal, please close this issue or leave a comment explaining why.

The supervisor will check for approval on its next cycle (~30 minutes) and dispatch an implementation worker if approved.

Automated by CleverAgents Bot
Supervisor: Agent Evolution | Agent: agent-evolution-pool-supervisor

⚠️ **Awaiting Human Approval** — This proposal requires human review before implementation. This issue was created by the Agent Evolution Supervisor (AUTO-EVLV) as a **Step 1 Proposal**. No code changes have been made yet. **To approve this proposal**, please do one of the following: 1. Remove the `needs feedback` label from this issue 2. Add the `State/Verified` label 3. Leave a comment with explicit approval (e.g., "approved" or "LGTM") **To reject this proposal**, please close this issue or leave a comment explaining why. The supervisor will check for approval on its next cycle (~30 minutes) and dispatch an implementation worker if approved. --- **Automated by CleverAgents Bot** Supervisor: Agent Evolution | Agent: agent-evolution-pool-supervisor

HAL9000 referenced this issue

2026-04-13 18:52:12 +00:00

[AUTO-EVLV] Status: Cycle 1 — Initial Agent Evolution Analysis (Cycle 1) #8398

HAL9000 referenced this issue

2026-04-13 19:25:59 +00:00

[AUTO-EVLV] Status: Cycle 1 — Initial Agent Evolution Analysis (Cycle 1) #8398