81b9f0f9ac2aea9c08e11f05b28153af7d850e0a

Message

Remove self-improvement loop from opencode

Diff

This diff is truncated to protect this page.

  1diff --git a/dot_config/opencode/AGENTS.md b/dot_config/opencode/AGENTS.md
  2index aa29863842d45f6d23a72c6a3a1a335d4f0096c2..72e8d32e07b1371912461accb206e1e30bb8219c 100644
  3--- a/dot_config/opencode/AGENTS.md
  4+++ b/dot_config/opencode/AGENTS.md
  5@@ -4,13 +4,6 @@ I'm a software engineer and hacker with preference for Linux, Go, UNIX philosoph
  6 
  7 Plan for context limits and session boundaries. If task has >2 substantial steps, the later will not be done in this session
  8 
  9-## Self-Improvement Loop
 10-
 11-- After ANY correction from the user: append to `~/.config/opencode/lessons.md` under the Raw section with `[YYYY-MM-DD]` timestamp, the pattern, and a preventive rule
 12-- Write rules for yourself that prevent the same mistake
 13-- Compaction and generalization happen autonomously via the `reflect` skill at session start
 14-- The user can also trigger `/reflect` for a collaborative session review
 15-
 16 ## Verification Before Done
 17 
 18 - Never mark a task complete without proving it works
 19diff --git a/dot_config/opencode/opencode.json b/dot_config/opencode/opencode.json
 20index 91b8db709173054bda5836b097b1f1caa4dec9ff..241ee6f47b1de24d98c69a29b8fb3775ad201f0c 100644
 21--- a/dot_config/opencode/opencode.json
 22+++ b/dot_config/opencode/opencode.json
 23@@ -1,100 +1,100 @@
 24 {
 25-	"$schema": "https://opencode.ai/config.json",
 26-	"autoupdate": false,
 27-	"permission": {
 28-		"edit": "allow",
 29-		"bash": {
 30-			"*": "ask",
 31-			"*sudo*": "deny",
 32-			"*terraform apply*": "deny"
 33-		},
 34-		"external_directory": {
 35-			"~/dev/dune/**": "allow",
 36-			"~/dev/go/**": "allow",
 37-			"*": "ask"
 38-		},
 39-		"webfetch": "ask",
 40-		"websearch": "allow",
 41-		"codesearch": "allow",
 42-		"doom_loop": "ask"
 43-	},
 44-	"plugin": [
 45-		"@tarquinen/opencode-dcp@latest",
 46-		"@theedgeofrage/opencode-telegram@0.2.2"
 47-	],
 48-	"provider": {
 49-		"llama.cpp": {
 50-			"npm": "@ai-sdk/openai-compatible",
 51-			"name": "llama-server (local)",
 52-			"options": {
 53-				"baseURL": "http://127.0.0.1:8080/v1"
 54-			},
 55-			"models": {
 56-				"qwen3.5-27b": {
 57-					"name": "Qwen3.5 27B (local)",
 58-					"limit": {
 59-						"context": 262144,
 60-						"output": 32768
 61-					}
 62-				}
 63-			}
 64-		}
 65-	},
 66-	"mcp": {
 67-		"Linear": {
 68-			"enabled": false,
 69-			"type": "remote",
 70-			"url": "https://mcp.linear.app/mcp",
 71-			"oauth": {}
 72-		},
 73-		"Sentry": {
 74-			"enabled": false,
 75-			"type": "remote",
 76-			"url": "https://mcp.sentry.dev/mcp",
 77-			"oauth": {}
 78-		},
 79-		"Grafana prod": {
 80-			"enabled": false,
 81-			"type": "local",
 82-			"command": ["uvx", "mcp-grafana"],
 83-			"environment": {
 84-				"GRAFANA_URL": "https://grafana.prod.internal.dunetech.io/",
 85-				"GRAFANA_SERVICE_ACCOUNT_TOKEN": "{file:~/hfs/mcp/grafana-prod-token}"
 86-			}
 87-		},
 88-		"Grafana dev": {
 89-			"enabled": false,
 90-			"type": "local",
 91-			"command": ["uvx", "mcp-grafana"],
 92-			"environment": {
 93-				"GRAFANA_URL": "https://grafana.dev.internal.dunetech.io/",
 94-				"GRAFANA_SERVICE_ACCOUNT_TOKEN": "{file:~/hfs/mcp/grafana-dev-token}"
 95-			}
 96-		},
 97-		"GitHub": {
 98-			"enabled": false,
 99-			"type": "remote",
100-			"url": "https://api.githubcopilot.com/mcp/",
101-			"headers": {
102-				"Authorization": "{file:~/hfs/mcp/github}"
103-			}
104-		},
105-		"Notion": {
106-			"enabled": false,
107-			"type": "remote",
108-			"url": "https://mcp.notion.com/mcp",
109-			"oauth": {}
110-		},
111-		"Dune": {
112-			"enabled": false,
113-			"type": "remote",
114-			"url": "https://api.dune.com/mcp/v1",
115-			"oauth": {}
116-		}
117-	},
118-	"formatter": {
119-		"clang-format": {
120-			"disabled": true
121-		}
122-	}
123diff --git a/dot_config/opencode/skills/weekly-review/SKILL.md b/dot_config/opencode/skills/weekly-review/SKILL.md
124deleted file mode 100644
125index 392dc89193dcdb582c3508c2767786e77632013b..0000000000000000000000000000000000000000
126--- a/dot_config/opencode/skills/weekly-review/SKILL.md
127+++ /dev/null
128@@ -1,115 +0,0 @@
129----
130-name: weekly-review
131-description: Analyze recent sessions to find patterns, recurring mistakes, and propose improvements to AGENTS.md and lessons.md.
132-compatibility: opencode
133----
134-
135-## What this is
136-
137-A periodic deep review of how the user and assistant work together. Reads session
138-history from the database, dispatches parallel agents to analyze conversations,
139-then runs multiple reflection loops to surface patterns and propose changes.
140-
141-## When to trigger
142-
143-User runs `/review-week`. Not autonomous — this is a deliberate review.
144-
145-## Inputs
146-
147-- `$ARGUMENTS`: optional time range in days (default: 7). Example: `/review-week 14`
148-
149-## Data source
150-
151-Sessions are stored in SQLite at `~/.local/share/opencode/opencode.db`.
152-
153-Key tables:
154-- `session`: metadata (id, title, directory, time_created, time_updated)
155-- `message`: per-session messages (id, session_id, data JSON)
156-- `part`: message content (id, message_id, session_id, data JSON with `type` and `text` fields)
157-
158-## Workflow
159-
160-### 1. Gather sessions
161-
162-```sql
163-SELECT id, title, directory,
164-  datetime(time_created/1000, 'unixepoch', 'localtime') as created,
165-  datetime(time_updated/1000, 'unixepoch', 'localtime') as updated
166-FROM session
167-WHERE time_created > (strftime('%s', 'now', '-N days') * 1000)
168-  AND title NOT LIKE '%@explore%'
169-  AND title NOT LIKE '%@general%'
170-ORDER BY time_created ASC;
171-```
172-
173-Replace `N` with the requested day range. Filter out subagent sessions — they're
174-noise for pattern analysis.
175-
176-Count messages per session to find the substantive ones (>3 messages).
177-
178-### 2. Dispatch parallel agents
179-
180-Group substantive sessions into 3-5 batches. For each batch, launch a `general`
181-subagent with instructions to:
182-
183-1. Extract conversation text from the `part` table:
184-   ```sql
185-   SELECT p.data FROM part p
186-   JOIN message m ON m.id = p.message_id
187-   WHERE m.session_id = 'SESSION_ID'
188-     AND json_extract(p.data, '$.type') = 'text'
189-   ORDER BY p.time_created ASC;
190-   ```
191-2. For each session, identify:
192-   - What the user was trying to accomplish
193-   - Corrections the user made to the assistant
194-   - Frustration signals
195-   - Where the assistant over- or under-delivered
196-   - Design decisions and their quality
197-3. Return a comprehensive analysis focused on patterns and lessons, not summaries.
198-
199-### 3. Read context files
200-
201-While agents run, read:
202-- `~/.config/opencode/AGENTS.md`
203-- `~/.config/opencode/lessons.md`
204-
205-### 4. Reflection loops
206-
207-Run 2-3 passes over the combined agent output:
208-
209-**Loop 1 — Patterns:** What themes repeat across sessions? Which are the most
210-frequent and most damaging? Cross-reference against existing AGENTS.md and
211-lessons.md rules — are these known problems (compliance failure) or new gaps?
212-
213-**Loop 2 — Design problems:** Why does the system produce these failures?
214-What's the root cause behind the patterns? Are existing rules too vague,
215-misplaced, or structurally unenforceable?
216-
217-**Loop 3 — Proposals:** Concrete, minimal changes. Prefer strengthening existing
218-rules over adding new ones. Every proposed change must map to a specific pattern
219-found in the data.
220-
221-### 5. Present findings
222-
223-Show the user:
224-- Session count and coverage
225-- Top patterns with frequency and severity
226-- The compliance question: which patterns already have rules?
227-- Proposed changes as exact edits (old → new)