Stated Intention Follow-Through

Rule

When you state "I'll do X" in a GitHub comment, journal, or other communication, you MUST show evidence of completion — either a commit, a PR, or a follow-up comment with a link to the work.

Context

When operating in agentic contexts with memory resets between sessions, stated intentions easily fall through the cracks. An agent writes "I'll create a lesson about this" in a comment, then the next session has no memory of the commitment. The requestor is left waiting with no idea whether the work happened.

Detection

Observable signals of this failure mode:

• Writing "I'll use X method going forward" with no commit showing the change
• Saying "I'll create a lesson about this" with no PR or issue tracking it
• Stating "I'll fix this" without linking to the fix
• Committing to a process change without documenting it anywhere
• Writing "I'll do X" in a comment, then moving on without evidence

Pattern

Every stated intention needs traceable follow-through:

Step 1: State Your Intention

I'll create a process document for this pattern.

Step 2: Execute the Work

Actually do the work you said you'd do.

Step 3: Show Evidence (CRITICAL)

## ✅ Process Document Created

I've created the process document as promised:

**Document**: [Link to file/commit/PR]
**Status**: Committed and pushed

Key elements:
- [Brief summary of what was done]
- [Why it addresses the original need]

Ready for review.

Evidence Types (Use One)

1. Commit link: "Committed in abc1234"
2. PR link: "PR #123 created: [URL]"
3. File link: "Created: [GitHub file URL]"
4. Issue comment: "Responded in same thread with results"

Anti-Patterns

❌ Silent Stated Intentions

"I'll use the log directory for accurate model tracking going forward."

[No commit, no PR, no follow-up]
[Next session: No evidence of change]

Problem: No accountability, no traceability, stakeholder has to follow up.

❌ Wrong-Arena Follow-Through

"I'll create a lesson for this."
[Creates file in personal notes, not committed]

"I'll fix this process."
[Documents in journal, not in shared knowledge base]

Problem: Work done but not discoverable by others or future sessions.

✅ Correct Pattern

"I'll create a process document for model verification."

[Creates file]
[Commits with descriptive message]
[Pushes to origin]

"✅ Created: knowledge/processes/model-verification-source-of-truth.md
Committed in 3fdeaa0.

Key content:
- Source of truth: ~/.local/share/gptme/logs/*/config.toml
- Why journals are unreliable for model tracking
- When to verify model usage"

Why This Matters

1. Trust: Stakeholders know commitments are kept
2. Traceability: Future sessions can find the work
3. Accountability: Clear record of what was done
4. Efficiency: No time wasted checking if things got done
5. Collaboration: Others can build on your work

Recovery Pattern

If you discover an unfulfilled stated intention:

1. Acknowledge immediately: "I said I'd do X but didn't show evidence"
2. Complete the work: Actually do what you promised
3. Show evidence: Link to commit/PR/file
4. Update lessons: Add this pattern to prevent repetition

Related

• Communication Loop Closure Patterns - Closing loops when completing requested actions
• Memory Failure Prevention - Broader context preservation
• Git Workflow - Proper commit practices

Origin

2026-02-13: Extracted from maintainer feedback on agent PRs where agents wrote "I'll do this" in comments but left no evidence of follow-through. The session-reset nature of LLM agents makes this failure mode systematic rather than incidental — without explicit evidence requirements, intentions evaporate between sessions.