Skip to content

feat: split session time into human hours vs machine hours #3921

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
marcusquinn merged 3 commits into from
Mar 8, 2026
Merged

feat: split session time into human hours vs machine hours #3921

Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
94a7aff
feat: split session time into human hours vs machine hours
marcusquinn Mar 8, 2026
3d0f452
fix: use total_human_hours from per-repo JSON in cross-repo aggregation
marcusquinn Mar 8, 2026
a4b2118
fix: add worker_human_hours to JSON schema and quarter to cross-repo …
marcusquinn Mar 8, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
196 changes: 123 additions & 73 deletions .agents/scripts/contributor-activity-helper.sh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -431,17 +431,19 @@
# Queries the OpenCode/Claude Code SQLite database to compute time spent
# in interactive sessions vs headless worker/runner sessions, per repo.
#
# Measures ACTUAL human time vs machine time per session using message
# timestamps: human_time = gap between assistant completing and next user
# message (reading + thinking + typing). machine_time = gap between
# assistant message created and completed (AI generating).
#
# Session type classification (by title pattern):
# - Worker: "Issue #*", "Supervisor Pulse", contains "/full-loop"
# - Worker: "Issue #*", "PR #*", "Supervisor Pulse", "/full-loop", "dispatch:", "Worker:"
# - Interactive: everything else (root sessions only)
# - Subagent: sessions with parent_id (excluded — time attributed to parent)
#
# Duration: max(message.time_created) - min(message.time_created) per session
# (actual active time between first and last message, not wall clock).
#
# Arguments:
# $1 - repo path (filters sessions by directory)
# --period day|week|month|year (optional, default: month)
# --period day|week|month|quarter|year (optional, default: month)
# --format markdown|json (optional, default: markdown)
# --db-path <path> (optional, default: auto-detect)
# Output: markdown table or JSON
Expand Down Expand Up @@ -486,7 +488,7 @@
db_path="${HOME}/.local/share/claude/Claude.db"
else
if [[ "$format" == "json" ]]; then
echo '{"interactive_sessions":0,"interactive_hours":0,"worker_sessions":0,"worker_hours":0}'
echo '{"interactive_sessions":0,"interactive_human_hours":0,"interactive_machine_hours":0,"worker_sessions":0,"worker_machine_hours":0,"total_human_hours":0,"total_machine_hours":0,"total_sessions":0}'
else
echo "_Session database not found._"
fi
Expand All @@ -496,7 +498,7 @@

if ! command -v sqlite3 &>/dev/null; then
if [[ "$format" == "json" ]]; then
echo '{"interactive_sessions":0,"interactive_hours":0,"worker_sessions":0,"worker_hours":0}'
echo '{"interactive_sessions":0,"interactive_human_hours":0,"interactive_machine_hours":0,"worker_sessions":0,"worker_machine_hours":0,"total_human_hours":0,"total_machine_hours":0,"total_sessions":0}'
else
echo "_sqlite3 not available._"
fi
Expand All @@ -509,6 +511,7 @@
day) seconds=86400 ;;
week) seconds=604800 ;;
month) seconds=2592000 ;;
quarter) seconds=7776000 ;;
year) seconds=31536000 ;;
*) seconds=2592000 ;;
esac
Expand All @@ -527,27 +530,52 @@
local like_path="${safe_path//%/\\%}"
like_path="${like_path//_/\\_}"

# Query session data with message-based duration using JSON output.
# JSON avoids pipe-separator issues (session titles can contain '|').
# Filters: root sessions only (no parent_id), within period, matching directory.
# Worktree directories (e.g., ~/Git/aidevops.feature-foo) are matched by prefix.
# Uses m.time_created for the period filter so sessions with recent messages
# are included even if the session itself was created before the cutoff.
# Query per-session human vs machine time using window functions.
# LAG() compares each message with the previous one in the same session:
# human_time = user.created - prev_assistant.completed (reading + thinking + typing)
# machine_time = assistant.completed - assistant.created (AI generating)
# Caps human gaps at 1 hour to exclude idle/abandoned sessions.
# Worker sessions (headless) have ~0% human time; interactive ~70-85%.
local query_result
query_result=$(sqlite3 -json "$db_path" "
WITH msg_data AS (
SELECT
s.id AS session_id,
s.title,
json_extract(m.data, '\$.role') AS role,
m.time_created AS created,
json_extract(m.data, '\$.time.completed') AS completed,
LAG(json_extract(m.data, '\$.role'))
OVER (PARTITION BY s.id ORDER BY m.time_created) AS prev_role,
LAG(json_extract(m.data, '\$.time.completed'))
OVER (PARTITION BY s.id ORDER BY m.time_created) AS prev_completed
FROM session s
JOIN message m ON m.session_id = s.id
WHERE s.parent_id IS NULL
AND m.time_created > ${since_ms}
AND (s.directory = '${safe_path}'
OR s.directory LIKE '${like_path}.%' ESCAPE '\\'
OR s.directory LIKE '${like_path}-%' ESCAPE '\\')
)
SELECT
s.title,
(max(m.time_created) - min(m.time_created)) as duration_ms
FROM session s
JOIN message m ON m.session_id = s.id
WHERE s.parent_id IS NULL
AND m.time_created > ${since_ms}
AND (s.directory = '${safe_path}'
OR s.directory LIKE '${like_path}.%' ESCAPE '\\'
OR s.directory LIKE '${like_path}-%' ESCAPE '\\')
GROUP BY s.id
HAVING count(m.id) >= 2
AND duration_ms > 5000
session_id,
title,
SUM(CASE
WHEN role = 'user' AND prev_role = 'assistant'
AND prev_completed IS NOT NULL
AND (created - prev_completed) BETWEEN 1 AND 3600000
THEN created - prev_completed
ELSE 0
END) AS human_ms,
SUM(CASE
WHEN role = 'assistant' AND completed IS NOT NULL
AND (completed - created) > 0
THEN completed - created
ELSE 0
END) AS machine_ms
FROM msg_data
GROUP BY session_id
HAVING human_ms + machine_ms > 5000
") || query_result="[]"
Comment on lines 540 to 579
Copy link
Copy Markdown

@gemini-code-assist gemini-code-assist bot Mar 8, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

security-high high

The SQL query on line 540 (in the session_time function) is constructed by interpolating shell variables ${since_ms}, ${safe_path}, and ${like_path} directly into the query string. This direct interpolation is vulnerable to SQL injection if an attacker can control the input (e.g., the repo path argument), potentially allowing arbitrary SQL commands against the Claude.db database. To prevent SQL injection in shell scripts using sqlite3, it is recommended to create a helper function that uses .param set for safe parameterized bindings instead of direct string interpolation. This approach is more robust than manual escaping.

References
  1. To prevent SQL injection in shell scripts using sqlite3, create a helper function that uses .param set for safe parameterized bindings instead of direct string interpolation.


# Process JSON in Python for classification and aggregation
Expand All @@ -562,6 +590,7 @@
# Worker session title patterns
worker_patterns = [
re.compile(r'^Issue #\d+'),
re.compile(r'^PR #\d+'),
re.compile(r'^Supervisor Pulse'),
re.compile(r'/full-loop', re.IGNORECASE),
re.compile(r'^dispatch:', re.IGNORECASE),
Expand All @@ -576,47 +605,56 @@

sessions = json.load(sys.stdin)

interactive_ms = 0
worker_ms = 0
interactive_count = 0
worker_count = 0
stats = {
'interactive': {'count': 0, 'human_ms': 0, 'machine_ms': 0},
'worker': {'count': 0, 'human_ms': 0, 'machine_ms': 0},
}

for row in sessions:
title = row.get('title', '')
duration_ms = row.get('duration_ms', 0)

session_type = classify_session(title)
if session_type == 'worker':
worker_ms += duration_ms
worker_count += 1
else:
interactive_ms += duration_ms
interactive_count += 1

interactive_hours = round(interactive_ms / 1000 / 3600, 1)
worker_hours = round(worker_ms / 1000 / 3600, 1)
total_hours = round((interactive_ms + worker_ms) / 1000 / 3600, 1)
human_ms = row.get('human_ms', 0)
machine_ms = row.get('machine_ms', 0)
stype = classify_session(title)
stats[stype]['count'] += 1
stats[stype]['human_ms'] += human_ms
stats[stype]['machine_ms'] += machine_ms

def ms_to_h(ms):
return round(ms / 3600000, 1)

i = stats['interactive']
w = stats['worker']
i_human_h = ms_to_h(i['human_ms'])
i_machine_h = ms_to_h(i['machine_ms'])
w_human_h = ms_to_h(w['human_ms'])
w_machine_h = ms_to_h(w['machine_ms'])
total_human_h = ms_to_h(i['human_ms'] + w['human_ms'])
total_machine_h = ms_to_h(i['machine_ms'] + w['machine_ms'])
total_sessions = i['count'] + w['count']

result = {
'interactive_hours': interactive_hours,
'interactive_sessions': interactive_count,
'worker_hours': worker_hours,
'worker_sessions': worker_count,
'total_hours': total_hours,
'total_sessions': interactive_count + worker_count,
'interactive_sessions': i['count'],
'interactive_human_hours': i_human_h,
'interactive_machine_hours': i_machine_h,
'worker_sessions': w['count'],
'worker_human_hours': w_human_h,
'worker_machine_hours': w_machine_h,
'total_human_hours': total_human_h,
'total_machine_hours': total_machine_h,
'total_sessions': total_sessions,
}

if format_type == 'json':
print(json.dumps(result, indent=2))
else:
if interactive_count == 0 and worker_count == 0:
if total_sessions == 0:
print(f'_No session data for the last {period_name}._')
else:
print(f'| Type | Sessions | Hours |')
print(f'| --- | ---: | ---: |')
print(f'| Interactive (human) | {interactive_count} | {interactive_hours}h |')
print(f'| Workers/Runners | {worker_count} | {worker_hours}h |')
print(f'| **Total** | **{interactive_count + worker_count}** | **{total_hours}h** |')
print(f'| Type | Sessions | Human Hours | Machine Hours |')
print(f'| --- | ---: | ---: | ---: |')
print(f'| Interactive | {i[\"count\"]} | {i_human_h}h | {i_machine_h}h |')
print(f'| Workers/Runners | {w[\"count\"]} | — | {w_machine_h}h |')
Comment on lines +655 to +656
Copy link
Copy Markdown

@gemini-code-assist gemini-code-assist bot Mar 8, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

medium

The backslashes (\) used to escape the double quotes within these f-strings are unnecessary. Since the f-strings are delimited by single quotes ('...'), the inner double quotes for dictionary key access (["count"]) are treated as literal characters by Python. The shell also passes them through correctly because the entire Python script is wrapped in double quotes. Removing these redundant escapes improves code readability.

Suggested change
print(f'| Interactive | {i[\"count\"]} | {i_human_h}h | {i_machine_h}h |')
print(f'| Workers/Runners | {w[\"count\"]} | — | {w_machine_h}h |')
print(f'| Interactive | {i["count"]} | {i_human_h}h | {i_machine_h}h |')
print(f'| Workers/Runners | {w["count"]} | — | {w_machine_h}h |')

print(f'| **Total** | **{total_sessions}** | **{total_human_h}h** | **{total_machine_h}h** |')
" "$format" "$period"

return 0
Expand All @@ -629,7 +667,7 @@
#
# Arguments:
# $1..N - repo paths
# --period day|week|month|year (optional, default: month)
# --period day|week|month|quarter|year (optional, default: month)
# --format markdown|json (optional, default: markdown)
# Output: aggregated table to stdout
#######################################
Expand Down Expand Up @@ -691,25 +729,34 @@
repos = json.load(sys.stdin)

totals = {
'interactive_hours': 0,
'interactive_sessions': 0,
'worker_hours': 0,
'interactive_human_hours': 0,
'interactive_machine_hours': 0,
'worker_sessions': 0,
'worker_human_hours': 0,
'worker_machine_hours': 0,
'total_human_hours': 0,
}

for repo in repos:
totals['interactive_hours'] += repo.get('interactive_hours', 0)
totals['interactive_sessions'] += repo.get('interactive_sessions', 0)
totals['worker_hours'] += repo.get('worker_hours', 0)
totals['interactive_human_hours'] += repo.get('interactive_human_hours', 0)
totals['interactive_machine_hours'] += repo.get('interactive_machine_hours', 0)
totals['worker_sessions'] += repo.get('worker_sessions', 0)
totals['worker_human_hours'] += repo.get('worker_human_hours', 0)
totals['worker_machine_hours'] += repo.get('worker_machine_hours', 0)
totals['total_human_hours'] += repo.get('total_human_hours', 0)

for k in ['interactive_human_hours', 'interactive_machine_hours', 'worker_human_hours', 'worker_machine_hours', 'total_human_hours']:
totals[k] = round(totals[k], 1)

totals['interactive_hours'] = round(totals['interactive_hours'], 1)
totals['worker_hours'] = round(totals['worker_hours'], 1)
total_hours = round(totals['interactive_hours'] + totals['worker_hours'], 1)
total_human_h = totals['total_human_hours']
total_machine_h = round(totals['interactive_machine_hours'] + totals['worker_machine_hours'], 1)
total_sessions = totals['interactive_sessions'] + totals['worker_sessions']

if format_type == 'json':
totals['total_hours'] = total_hours
totals['total_human_hours'] = total_human_h
totals['total_machine_hours'] = total_machine_h
totals['total_sessions'] = total_sessions
totals['repo_count'] = repo_count
print(json.dumps(totals, indent=2))
Expand All @@ -719,11 +766,12 @@
else:
print(f'_Across {repo_count} managed repos:_')
print()
print(f'| Type | Sessions | Hours |')
print(f'| --- | ---: | ---: |')
print(f'| Interactive (human) | {totals[\"interactive_sessions\"]} | {totals[\"interactive_hours\"]}h |')
print(f'| Workers/Runners | {totals[\"worker_sessions\"]} | {totals[\"worker_hours\"]}h |')
print(f'| **Total** | **{total_sessions}** | **{total_hours}h** |')
print(f'| Type | Sessions | Human Hours | Machine Hours |')
print(f'| --- | ---: | ---: | ---: |')
i = totals
print(f'| Interactive | {i[\"interactive_sessions\"]} | {i[\"interactive_human_hours\"]}h | {i[\"interactive_machine_hours\"]}h |')
print(f'| Workers/Runners | {i[\"worker_sessions\"]} | — | {i[\"worker_machine_hours\"]}h |')
Comment on lines +772 to +773
Copy link
Copy Markdown

@gemini-code-assist gemini-code-assist bot Mar 8, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

medium

Similar to the other comment, the backslashes (\) for escaping double quotes here are redundant. The f-strings use single quotes, so the inner double quotes for dictionary keys don't need to be escaped for either the shell or Python. Removing them will make the code cleaner and easier to maintain.

Suggested change
print(f'| Interactive | {i[\"interactive_sessions\"]} | {i[\"interactive_human_hours\"]}h | {i[\"interactive_machine_hours\"]}h |')
print(f'| Workers/Runners | {i[\"worker_sessions\"]} | — | {i[\"worker_machine_hours\"]}h |')
print(f'| Interactive | {i["interactive_sessions"]} | {i["interactive_human_hours"]}h | {i["interactive_machine_hours"]}h |')
print(f'| Workers/Runners | {i["worker_sessions"]} | — | {i["worker_machine_hours"]}h |')

print(f'| **Total** | **{total_sessions}** | **{total_human_h}h** | **{total_machine_h}h** |')
" "$format" "$period" "$repo_count"

return 0
Expand All @@ -741,7 +789,7 @@
local repo_path="${1:-.}"
shift || true
local period="month"
local format="markdown"

Check warning on line 792 in .agents/scripts/contributor-activity-helper.sh

View check run for this annotation

SonarQubeCloud / SonarCloud Code Analysis

Define a constant instead of using the literal 'markdown' 4 times. 


See more on https://sonarcloud.io/project/issues?id=marcusquinn_aidevops&issues=AZzPMrQKTBjk2HgQCcxu&open=AZzPMrQKTBjk2HgQCcxu&pullRequest=3921
while [[ $# -gt 0 ]]; do
case "$1" in
--period)
Expand Down Expand Up @@ -785,8 +833,8 @@
echo " table <repo-path> [--period day|week|month|year] [--format markdown|json]"
echo " user <repo-path> <github-login>"
echo " cross-repo-summary <path1> [path2 ...] [--period month] [--format markdown]"
echo " session-time <repo-path> [--period month] [--format markdown]"
echo " cross-repo-session-time <path1> [path2 ...] [--period month] [--format markdown]"
echo " session-time <repo-path> [--period day|week|month|quarter|year] [--format markdown|json]"
echo " cross-repo-session-time <path1> [path2 ...] [--period month] [--format markdown|json]"
echo ""
echo "Computes contributor activity from immutable git commit history."
echo "Session time stats from AI assistant database (OpenCode/Claude Code)."
Expand All @@ -796,9 +844,11 @@
echo " Direct - committer is the author (push, CLI commit)"
echo " PR Merge - committer is noreply@github.com (GitHub squash-merge)"
echo ""
echo "Session types:"
echo " Interactive - human-driven sessions (conversations, debugging)"
echo " Worker - headless dispatched tasks (Issue #N, Supervisor Pulse)"
echo "Session time (human vs machine):"
echo " Human hours - time spent reading, thinking, typing (between AI responses)"
echo " Machine hours - time AI spent generating responses"
echo " Interactive - human-driven sessions (conversations, debugging)"
echo " Worker - headless dispatched tasks (Issue #N, PR #N, Supervisor Pulse)"
return 0
;;
esac
Expand Down